Merge pull request 'feat(fb-30): MCP server (stdio) — agent integration MVP' (#108) from feat/p9-fb-30-mcp-server into main

Reviewed-on: #108
This commit was merged in pull request #108.
This commit is contained in:
2026-05-07 08:04:53 +00:00
36 changed files with 1207 additions and 20 deletions

View File

@@ -4,7 +4,7 @@ This file provides guidance to Claude Code (claude.ai/code) when working with co
## Project
Single-user local-first knowledge base + RAG. Rust 2024 workspace, ~20 crates, single binary (`kebab`). All inference is local (Ollama + fastembed + whisper.cpp).
Single-user local-first knowledge base + RAG. Rust 2024 workspace, ~21 crates, single binary (`kebab`). All inference is local (Ollama + fastembed + whisper.cpp).
The repo's documentation is split by audience — don't duplicate across them:
@@ -54,7 +54,7 @@ Each task spec lists `Allowed dependencies` and `Forbidden dependencies` per des
- `kebab-core` MUST NOT depend on any other `kebab-*` crate. Domain types only.
- `kebab-eval`'s `metrics` and `compare` modules MUST NOT import retrieval / embedding / LLM crates directly. The runner is allowed to use `kebab-app`'s facade (P5-1 inheritance — see deviations in that task spec).
- UI crates (`kebab-cli`, future `kebab-tui`, `kebab-desktop`) MUST NOT import `kebab-store-*` / `kebab-llm-*` / `kebab-parse-*` directly — only `kebab-app`.
- UI crates (`kebab-cli`, `kebab-mcp`, `kebab-tui`, future `kebab-desktop`) MUST NOT import `kebab-store-*` / `kebab-llm-*` / `kebab-parse-*` directly — only `kebab-app`.
Read the relevant task spec's deps section before adding an import. New crates inherit the same boundary rules.

121
Cargo.lock generated
View File

@@ -549,7 +549,7 @@ dependencies = [
"log",
"num-rational",
"num-traits",
"pastey",
"pastey 0.1.1",
"rayon",
"thiserror 2.0.18",
"v_frame",
@@ -1229,6 +1229,16 @@ dependencies = [
"darling_macro 0.21.3",
]
[[package]]
name = "darling"
version = "0.23.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "25ae13da2f202d56bd7f91c25fba009e7717a1e4a1cc98a76d844b65ae912e9d"
dependencies = [
"darling_core 0.23.0",
"darling_macro 0.23.0",
]
[[package]]
name = "darling_core"
version = "0.20.11"
@@ -1257,6 +1267,19 @@ dependencies = [
"syn 2.0.117",
]
[[package]]
name = "darling_core"
version = "0.23.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "9865a50f7c335f53564bb694ef660825eb8610e0a53d3e11bf1b0d3df31e03b0"
dependencies = [
"ident_case",
"proc-macro2",
"quote",
"strsim",
"syn 2.0.117",
]
[[package]]
name = "darling_macro"
version = "0.20.11"
@@ -1279,6 +1302,17 @@ dependencies = [
"syn 2.0.117",
]
[[package]]
name = "darling_macro"
version = "0.23.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "ac3984ec7bd6cfa798e62b4a642426a5be0e68f9401cfc2a01e3fa9ea2fcdb8d"
dependencies = [
"darling_core 0.23.0",
"quote",
"syn 2.0.117",
]
[[package]]
name = "dary_heap"
version = "0.3.9"
@@ -3516,6 +3550,7 @@ dependencies = [
"kebab-store-vector",
"lopdf",
"lru",
"reqwest",
"rusqlite",
"serde",
"serde_json",
@@ -3557,8 +3592,8 @@ dependencies = [
"kebab-config",
"kebab-core",
"kebab-eval",
"kebab-mcp",
"kebab-tui",
"reqwest",
"serde",
"serde_json",
"tempfile",
@@ -3666,6 +3701,23 @@ dependencies = [
"wiremock",
]
[[package]]
name = "kebab-mcp"
version = "0.3.0"
dependencies = [
"anyhow",
"kebab-app",
"kebab-config",
"kebab-core",
"rmcp",
"schemars 1.2.1",
"serde",
"serde_json",
"tempfile",
"tokio",
"tracing",
]
[[package]]
name = "kebab-normalize"
version = "0.3.0"
@@ -5457,6 +5509,12 @@ version = "0.1.1"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "35fb2e5f958ec131621fdd531e9fc186ed768cbe395337403ae56c17a74c68ec"
[[package]]
name = "pastey"
version = "0.2.2"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "c5a797f0e07bdf071d15742978fc3128ec6c22891c31a3a931513263904c982a"
[[package]]
name = "path_abs"
version = "0.5.1"
@@ -6306,6 +6364,40 @@ dependencies = [
"windows-sys 0.52.0",
]
[[package]]
name = "rmcp"
version = "1.6.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "e12ca9067b5ebfbd5b3fcdc4acfceb81aa7d5ab2a879dff7cb75d22434276aad"
dependencies = [
"async-trait",
"chrono",
"futures",
"pastey 0.2.2",
"pin-project-lite",
"rmcp-macros",
"schemars 1.2.1",
"serde",
"serde_json",
"thiserror 2.0.18",
"tokio",
"tokio-util",
"tracing",
]
[[package]]
name = "rmcp-macros"
version = "1.6.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "7caa6743cc0888e433105fe1bc551a7f607940b126a37bc97b478e86064627eb"
dependencies = [
"darling 0.23.0",
"proc-macro2",
"quote",
"serde_json",
"syn 2.0.117",
]
[[package]]
name = "roaring"
version = "0.10.12"
@@ -6512,12 +6604,26 @@ version = "1.2.1"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "a2b42f36aa1cd011945615b92222f6bf73c599a102a300334cd7f8dbeec726cc"
dependencies = [
"chrono",
"dyn-clone",
"ref-cast",
"schemars_derive",
"serde",
"serde_json",
]
[[package]]
name = "schemars_derive"
version = "1.2.1"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "7d115b50f4aaeea07e79c1912f645c7513d81715d0420f8bc77a18c6260b307f"
dependencies = [
"proc-macro2",
"quote",
"serde_derive_internals",
"syn 2.0.117",
]
[[package]]
name = "scoped-tls"
version = "1.0.1"
@@ -6606,6 +6712,17 @@ dependencies = [
"syn 2.0.117",
]
[[package]]
name = "serde_derive_internals"
version = "0.29.1"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "18d26a20a969b9e3fdf2fc2d9f21eda6c40e2de84c9408bb5d3b05d499aae711"
dependencies = [
"proc-macro2",
"quote",
"syn 2.0.117",
]
[[package]]
name = "serde_json"
version = "1.0.149"

View File

@@ -22,6 +22,7 @@ members = [
"crates/kebab-parse-image",
"crates/kebab-parse-pdf",
"crates/kebab-tui",
"crates/kebab-mcp",
]
[workspace.package]
@@ -71,6 +72,10 @@ futures = "0.3"
# pass; pulled into the workspace deps so future crates can share the
# same major.
regex = "1"
# MCP (Model Context Protocol) SDK. server + macros + transport-io provide
# stdio JSON-RPC transport for `kebab-mcp` (p9-fb-30). schemars feature
# exposes the derive macro used by tool input schemas.
rmcp = { version = "1.6", default-features = false, features = ["server", "macros", "transport-io", "schemars"] }
# Dev-only HTTP mock server for kebab-llm-local Ollama adapter tests. Requires
# a tokio runtime to host its mock server (the runtime adapter crate stays
# sync via reqwest::blocking — wiremock is dev-only there).

View File

@@ -31,6 +31,7 @@ P0~P5 직렬. P6~P9 P5 이후 병렬 가능.
머지 후 발견된 모든 deviation / hotfix 의 dated 로그는 [tasks/HOTFIXES.md](tasks/HOTFIXES.md). 본 요약은 \"누군가가 인수받을 때 알아두면 시간을 많이 절약하는\" 항목만:
- **2026-05-07 P9 post-도그푸딩 (p9-fb-30)** — `kebab mcp` 신규 subcommand + new crate `kebab-mcp` (lib only) — stdio JSON-RPC server. 4 read-only tool (`search` / `ask` / `schema` / `doctor`) 가 `kebab-app` facade 위에 build. rmcp 1.6 SDK 채택, manual `tools/list` + `tools/call` dispatch (rmcp 의 `#[tool_router]` 매크로 대신). `error_classify` 모듈을 `kebab-cli``kebab-app::error_wire` 로 promotion (UI crate 끼리 import 회피, facade 룰 준수). `ErrorV1``schema_version: String` 필드 추가 — kebab-mcp 의 직접 serialize 경로에서도 wire 정합. `KebabAppState``(Config, Option<PathBuf>)` carry — doctor tool 의 path-aware behavior 위해. ask + search arm 의 `tokio::task::spawn_blocking` wrap — `OllamaLanguageModel` 의 reqwest blocking client 가 async 안에서 panic 회피. capability flag `mcp_server` `false``true`. agent integration MVP 완성 — Claude Code / Cursor / OpenAI Agents 등 host-agnostic 사용 가능. spec: `tasks/p9/p9-fb-30-mcp-server.md`. design: `docs/superpowers/specs/2026-05-07-p9-fb-30-mcp-server-design.md`.
- **P3-5 / P4-3 `--config` 누락** — `kebab-cli``--config <path>` 를 honor 하려면 `kebab_app::*_with_config` companion 을 호출해야 함. 두 번 같은 모양으로 회귀했음.
- **P6-2 OCR 기본 엔진** — spec literal 의 Tesseract 가 시스템 dep 부담으로 거부됨, Ollama vision LM 으로 대체. `OcrEngine` trait 그대로라 future swap 가능.
- **P6-3 caption** — `GenerateRequest.images` 필드를 `kebab-core::LanguageModel` trait 에 신설. 기존 caller 모두 `images: Vec::new()` 로 마이그레이션.

View File

@@ -80,6 +80,7 @@ kebab doctor
| `kebab reset [--all / --data-only / --vector-only / --config-only] [--yes]` | XDG 데이터 wipe. **Irreversible.** TTY 면 confirm prompt, 아니면 `--yes` 필수. `--vector-only` 는 SQLite `embedding_records` 도 함께 truncate (orphan 방지) |
| `kebab eval run / compare` | golden query 회귀 측정 |
| `kebab schema [--json]` | introspection — wire schemas / capabilities / models / stats 한 번에. `--json``schema.v1` wire; 사람 모드는 서식 출력. |
| `kebab mcp` | MCP (Model Context Protocol) stdio server. agent host (Claude Code / Cursor / OpenAI Agents) 가 spawn 하여 tool 호출 (`search` / `ask` / `schema` / `doctor`). `--config` honor. |
모든 명령에 `--json` 플래그. 출력은 frozen wire schema v1 (`schema_version` 항상 포함, 예: `ingest_report.v1`, `ingest_progress.v1`, `search_hit.v1`, `answer.v1`, `doctor.v1`, `reset_report.v1`, `schema.v1`). `--json` 모드에서 fatal error 는 stderr 에 `error.v1` ndjson 으로 emit (exit code 0/1/2/3 unchanged).
@@ -160,9 +161,26 @@ config 예시는 [docs/SMOKE.md](docs/SMOKE.md) 의 `/tmp/kebab-smoke/config.tom
- **Claude Code skill** — repo 의 [`integrations/claude-code/`](integrations/claude-code/) 가 ship-ready skill. `cp -r integrations/claude-code/kebab ~/.claude/skills/` 한 번이면 새 Claude Code 세션부터 자동 trigger (내부 시스템 / 위키 lookup / 사내 runbook 질문). multi-turn 은 `kebab ask --session <id> --json` 으로 영속 — skill 이 conversation id 관리하면 외부 agent 도 `--repl` 없이 stateful 대화 가능 (p9-fb-18).
- **Codex / 기타 agent host** — `--json` + frozen wire schema v1 가 stable contract. 동일 패턴으로 ~50줄 wrapper 작성 가능. `integrations/<host>/` 에 추가 PR 환영.
- **MCP server** — stdio JSON-RPC 로 `kebab-app` facade 1:1 노출.
- **MCP server** — stdio JSON-RPC 로 `kebab-app` facade 1:1 노출. `kebab mcp` 참조.
- **HTTP wrapper** — `kebab serve --bind 127.0.0.1:7711` (P+, local-only 가치 신중).
## MCP 사용 (Claude Code 예시)
`~/.claude/mcp.json` (또는 host 의 동등 위치):
```json
{
"mcpServers": {
"kebab": {
"command": "kebab",
"args": ["mcp"]
}
}
}
```
Claude Code 가 session 시작 시 `kebab mcp` 를 spawn — process 가 session 동안 살아 있어 SQLite / Lance / fastembed 가 hot. 4 tool: `search` (lexical/vector/hybrid 검색), `ask` (RAG 답변, optional `session_id` for multi-turn + optional `mode` override), `schema` (capability 조회), `doctor` (health check). 모든 tool 의 결과는 wire schema v1 JSON 으로 text content 안에 직렬화 — agent 가 parse 후 사용. tool dispatch 실패 (잘못된 config / 미초기화 KB 등) 는 `isError: true` + error.v1 content; refusal / no-hit / unhealthy 는 정상 응답 (semantic flag 으로 분기).
## 비-목표
다중 사용자 SaaS / K8s / 원격 vector DB / enterprise RBAC / 실시간 협업 / 모든 파일 포맷의 완벽한 parsing / agent 임의 파일 수정 / multi-workspace / LLM-as-judge eval / CLIP 시각 embedding / `kebab://` protocol handler — frozen 설계 §11 / §0 참조.

View File

@@ -64,3 +64,6 @@ image = { version = "0.25", default-features = false, features =
# to the same major (0.32) so byte output is identical between the two
# fixture surfaces.
lopdf = "0.32"
# error_wire::tests::llm_unreachable_classifies_to_model_unreachable needs a real
# reqwest::Error (private constructor) — built from a connect-refused call.
reqwest = { version = "0.12", default-features = false, features = ["blocking", "rustls-tls"] }

View File

@@ -9,10 +9,15 @@
use serde::{Deserialize, Serialize};
use serde_json::{Value, json};
use kebab_app::error_signal::{ConfigInvalid, LlmError, NotIndexed};
use crate::error_signal::{ConfigInvalid, LlmError, NotIndexed};
/// Wire schema id for [`ErrorV1`]. Single source of truth — kebab-cli
/// + kebab-mcp use this via `kebab_app::ERROR_V1_ID`.
pub const ERROR_V1_ID: &str = "error.v1";
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct ErrorV1 {
pub schema_version: String,
pub code: String,
pub message: String,
pub details: Value,
@@ -22,6 +27,7 @@ pub struct ErrorV1 {
pub fn classify(err: &anyhow::Error, verbose: bool) -> ErrorV1 {
if let Some(s) = err.downcast_ref::<ConfigInvalid>() {
return ErrorV1 {
schema_version: ERROR_V1_ID.to_string(),
code: "config_invalid".to_string(),
message: s.to_string(),
details: json!({
@@ -33,6 +39,7 @@ pub fn classify(err: &anyhow::Error, verbose: bool) -> ErrorV1 {
}
if let Some(s) = err.downcast_ref::<NotIndexed>() {
return ErrorV1 {
schema_version: ERROR_V1_ID.to_string(),
code: "not_indexed".to_string(),
message: s.to_string(),
details: json!({
@@ -47,6 +54,7 @@ pub fn classify(err: &anyhow::Error, verbose: bool) -> ErrorV1 {
}
if let Some(io) = err.downcast_ref::<std::io::Error>() {
return ErrorV1 {
schema_version: ERROR_V1_ID.to_string(),
code: "io_error".to_string(),
message: io.to_string(),
details: json!({"kind": format!("{:?}", io.kind())}),
@@ -59,6 +67,7 @@ pub fn classify(err: &anyhow::Error, verbose: bool) -> ErrorV1 {
details = json!({"chain": chain});
}
ErrorV1 {
schema_version: ERROR_V1_ID.to_string(),
code: "generic".to_string(),
message: err.to_string(),
details,
@@ -69,6 +78,7 @@ pub fn classify(err: &anyhow::Error, verbose: bool) -> ErrorV1 {
fn classify_llm(s: &LlmError) -> ErrorV1 {
match s {
LlmError::Unreachable { endpoint, source } => ErrorV1 {
schema_version: ERROR_V1_ID.to_string(),
code: "model_unreachable".to_string(),
message: format!("ollama unreachable at {endpoint}"),
details: json!({
@@ -78,24 +88,28 @@ fn classify_llm(s: &LlmError) -> ErrorV1 {
hint: Some(format!("ensure `ollama serve` is reachable at {endpoint}")),
},
LlmError::ModelNotPulled(model) => ErrorV1 {
schema_version: ERROR_V1_ID.to_string(),
code: "model_not_pulled".to_string(),
message: format!("ollama model `{model}` is not pulled"),
details: json!({"model": model}),
hint: Some(format!("run `ollama pull {model}`")),
},
LlmError::Timeout(e) => ErrorV1 {
schema_version: ERROR_V1_ID.to_string(),
code: "timeout".to_string(),
message: format!("ollama timeout: {e}"),
details: json!({"source": e.to_string()}),
hint: Some("increase timeout or check Ollama load".to_string()),
},
LlmError::Stream(body) => ErrorV1 {
schema_version: ERROR_V1_ID.to_string(),
code: "generic".to_string(),
message: format!("ollama HTTP error: {body}"),
details: json!({"body": body}),
hint: None,
},
LlmError::Malformed(line) => ErrorV1 {
schema_version: ERROR_V1_ID.to_string(),
code: "generic".to_string(),
message: format!("malformed response line: {line}"),
details: json!({"line": line}),

View File

@@ -57,6 +57,7 @@ use kebab_source_fs::FsSourceConnector;
mod app;
pub mod doctor_signal;
pub mod error_signal;
pub mod error_wire;
pub mod ingest_progress;
pub mod logging;
pub mod reset;
@@ -65,6 +66,7 @@ pub mod schema;
pub use app::App;
pub use ingest_progress::{AggregateCounts, IngestEvent, render_skipped_breakdown};
pub use reset::{ResetReport, ResetScope};
pub use error_wire::{ERROR_V1_ID, ErrorV1, classify};
pub use schema::{Capabilities, Models, SCHEMA_V1_ID, SchemaV1, Stats, WireBlock, schema_with_config};
/// p9-fb-25: sentinel for files without an extension in

View File

@@ -108,7 +108,7 @@ fn capabilities_snapshot() -> Capabilities {
incremental_ingest: true,
streaming_ask: false,
http_daemon: false,
mcp_server: false,
mcp_server: true,
single_file_ingest: false,
}
}

View File

@@ -58,6 +58,10 @@ fn schema_report_reflects_freshly_ingested_kb() {
);
assert!(schema.capabilities.json_mode);
assert!(!schema.capabilities.streaming_ask);
assert!(
schema.capabilities.mcp_server,
"mcp_server should be true after fb-30",
);
assert_eq!(
schema.stats.doc_count, 2,
"expected 2 docs (a.md + b.md): {:?}",

View File

@@ -27,6 +27,8 @@ kebab-eval = { path = "../kebab-eval" }
# enforces the §8 boundary in its own Cargo.toml; kb-cli just
# launches it.
kebab-tui = { path = "../kebab-tui" }
# p9-fb-30: MCP stdio server. `Cmd::Mcp` delegates entirely to this crate.
kebab-mcp = { path = "../kebab-mcp" }
anyhow = { workspace = true }
serde = { workspace = true }
serde_json = { workspace = true }
@@ -44,6 +46,3 @@ ctrlc = "3"
[dev-dependencies]
tempfile = { workspace = true }
# llm_unreachable_classifies_to_model_unreachable test needs a real
# reqwest::Error (private constructor) — built from a connect-refused call.
reqwest = { version = "0.12", default-features = false, features = ["blocking", "rustls-tls"] }

View File

@@ -9,7 +9,6 @@ use clap::{Parser, Subcommand};
use kebab_app::doctor_signal::{DoctorUnhealthy, NoHitSignal, RefusalSignal};
mod cancel;
mod error_classify;
mod progress;
mod wire;
@@ -189,6 +188,11 @@ enum Cmd {
#[command(subcommand)]
what: EvalWhat,
},
/// Run the MCP (Model Context Protocol) stdio server. Used by
/// agent hosts (Claude Code / Cursor / OpenAI Agents) to call kebab
/// tools (search / ask / schema / doctor).
Mcp,
}
#[derive(Subcommand, Debug)]
@@ -282,7 +286,7 @@ fn main() -> ExitCode {
// caller); errors go to stderr.
if code != 1 {
if cli.json {
let v1 = error_classify::classify(&e, cli.verbose);
let v1 = kebab_app::classify(&e, cli.verbose);
let v = wire::wire_error_v1(&v1);
eprintln!("{}", serde_json::to_string(&v).unwrap_or_else(|_| {
"{\"schema_version\":\"error.v1\",\"code\":\"generic\",\"message\":\"serialize failed\"}".to_string()
@@ -740,6 +744,11 @@ fn run(cli: &Cli) -> anyhow::Result<()> {
Ok(())
}
},
Cmd::Mcp => {
let cfg = kebab_config::Config::load(cli.config.as_deref())?;
kebab_mcp::serve_stdio(cfg, cli.config.clone())
}
}
}

View File

@@ -152,12 +152,12 @@ pub fn wire_schema(s: &kebab_app::SchemaV1) -> Value {
tag_object(v, kebab_app::SCHEMA_V1_ID)
}
/// Wrap an [`crate::error_classify::ErrorV1`] as `error.v1`.
/// Wrap an [`kebab_app::ErrorV1`] as `error.v1`.
///
/// Uses the simple `tag_object` pattern because `ErrorV1` is a
/// kebab-cli-local type that does NOT carry `schema_version` itself
/// type that does NOT carry `schema_version` itself
/// (kebab-core convention).
pub fn wire_error_v1(e: &crate::error_classify::ErrorV1) -> Value {
pub fn wire_error_v1(e: &kebab_app::ErrorV1) -> Value {
let v = serde_json::to_value(e).expect("ErrorV1 serializes");
tag_object(v, "error.v1")
}
@@ -262,8 +262,9 @@ mod tests {
#[test]
fn error_wrapper_tags_schema_version_and_emits_code() {
use crate::error_classify::ErrorV1;
use kebab_app::ErrorV1;
let err = ErrorV1 {
schema_version: "error.v1".to_string(),
code: "config_invalid".to_string(),
message: "bad config".to_string(),
details: serde_json::json!({"path": "/tmp/x"}),

View File

@@ -0,0 +1,77 @@
//! Spawn `target/debug/kebab mcp` and exercise initialize → tools/list.
//!
//! rmcp 1.6 has no public in-memory test transport, so this is the only
//! end-to-end MCP assertion in the suite. The binary is located via
//! `CARGO_BIN_EXE_kebab` which cargo injects at test compile time.
use std::io::{BufRead, BufReader, Write};
use std::process::{Command, Stdio};
#[test]
fn cli_mcp_initialize_then_tools_list() {
let bin = env!("CARGO_BIN_EXE_kebab");
let mut child = Command::new(bin)
.arg("mcp")
.stdin(Stdio::piped())
.stdout(Stdio::piped())
.stderr(Stdio::null())
.spawn()
.unwrap();
let mut stdin = child.stdin.take().unwrap();
let stdout = child.stdout.take().unwrap();
let mut reader = BufReader::new(stdout);
// rmcp 1.6 defaults to protocol version "2025-03-26" (confirmed by
// manual smoke in Task 10). The server echoes whatever version the
// client sends during the handshake, so this literal must match.
let init_req = r#"{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2025-03-26","capabilities":{},"clientInfo":{"name":"test","version":"0"}}}"#;
writeln!(stdin, "{init_req}").unwrap();
writeln!(
stdin,
r#"{{"jsonrpc":"2.0","method":"notifications/initialized"}}"#
)
.unwrap();
writeln!(
stdin,
r#"{{"jsonrpc":"2.0","id":2,"method":"tools/list","params":{{}}}}"#
)
.unwrap();
// Read initialize response.
let mut line = String::new();
reader.read_line(&mut line).unwrap();
let init: serde_json::Value = serde_json::from_str(line.trim()).unwrap();
assert_eq!(
init.get("id").and_then(|i| i.as_i64()),
Some(1),
"unexpected id in initialize response: {init}"
);
assert!(
init.get("result").is_some(),
"initialize result missing: {init}"
);
// Read tools/list response.
line.clear();
reader.read_line(&mut line).unwrap();
let list: serde_json::Value = serde_json::from_str(line.trim()).unwrap();
assert_eq!(
list.get("id").and_then(|i| i.as_i64()),
Some(2),
"unexpected id in tools/list response: {list}"
);
let tools = list["result"]["tools"]
.as_array()
.expect("tools/list result.tools must be an array");
assert_eq!(
tools.len(),
4,
"expected 4 tools (schema, doctor, search, ask), got {}: {list}",
tools.len()
);
// Gracefully close stdin so the server shuts down cleanly.
drop(stdin);
let _ = child.wait().unwrap();
}

View File

@@ -92,8 +92,8 @@ fn cli_schema_json_emits_schema_v1() {
);
assert_eq!(
caps.get("mcp_server").and_then(|b| b.as_bool()),
Some(false),
"capabilities.mcp_server must be false (not yet shipped)"
Some(true),
"capabilities.mcp_server must be true (fb-30)"
);
}

View File

@@ -0,0 +1,27 @@
[package]
name = "kebab-mcp"
edition = { workspace = true }
rust-version = { workspace = true }
license = { workspace = true }
repository = { workspace = true }
version = { workspace = true }
[dependencies]
rmcp = { workspace = true }
# rt-multi-thread + io-util + io-std extend the workspace tokio entry
# (which only declares rt + macros) for the blocking stdio MCP transport.
tokio = { workspace = true, features = ["rt-multi-thread", "macros", "io-util", "io-std"] }
serde = { workspace = true }
serde_json = { workspace = true }
anyhow = { workspace = true }
tracing = { workspace = true }
# schemars 1.x matches rmcp 1.6's ^1.0 requirement (verified via crates.io
# /dependencies endpoint — rmcp declares optional schemars = "^1.0").
schemars = "1"
kebab-app = { path = "../kebab-app" }
kebab-config = { path = "../kebab-config" }
kebab-core = { path = "../kebab-core" }
[dev-dependencies]
tempfile = { workspace = true }

View File

@@ -0,0 +1,22 @@
//! Map `anyhow::Error` returned by kebab-app facades to MCP
//! `CallToolResult` with `isError: true` + error.v1 JSON content.
use rmcp::model::{CallToolResult, Content};
use kebab_app::classify;
/// Convert an `anyhow::Error` to a `CallToolResult` with `isError: true`
/// and the serialised `error.v1` envelope as the text content.
pub fn to_tool_error(err: &anyhow::Error) -> CallToolResult {
let v1 = classify(err, false);
let body = serde_json::to_string(&v1).unwrap_or_else(|_| {
r#"{"schema_version":"error.v1","code":"generic","message":"serialize failed"}"#
.to_string()
});
CallToolResult::error(vec![Content::text(body)])
}
/// Wrap a successful wire-schema JSON string as a `CallToolResult`.
pub fn to_tool_success(json: String) -> CallToolResult {
CallToolResult::success(vec![Content::text(json)])
}

164
crates/kebab-mcp/src/lib.rs Normal file
View File

@@ -0,0 +1,164 @@
//! MCP (Model Context Protocol) server over stdio. Exposes 4 read-only
//! tools (`search` / `ask` / `schema` / `doctor`) backed by `kebab-app`
//! facade methods. Used by `kebab-cli`'s `Cmd::Mcp` arm.
//!
//! See spec `docs/superpowers/specs/2026-05-07-p9-fb-30-mcp-server-design.md`.
use std::path::PathBuf;
use anyhow::Result;
use rmcp::ServerHandler;
use rmcp::handler::server::common::{schema_for_empty_input, schema_for_type};
use rmcp::model::{
CallToolRequestParams, CallToolResult, Implementation, ListToolsResult, ServerCapabilities,
ServerInfo, Tool,
};
use rmcp::service::{RequestContext, ServiceExt};
use rmcp::transport::stdio;
use rmcp::{ErrorData, RoleServer};
use kebab_config::Config;
pub mod error;
pub mod state;
pub mod tools;
pub use state::KebabAppState;
/// Build the canonical list of tools exposed by the MCP server.
///
/// Extracted from [`ServerHandler::list_tools`] so it can be called
/// directly in tests without constructing a `RequestContext`.
pub fn build_tools_vec() -> Vec<Tool> {
vec![
Tool::new(
"schema",
"Introspection — wire schemas, capabilities, model versions, index stats.",
schema_for_empty_input(),
),
Tool::new(
"doctor",
"Health check — verifies config, storage, models, and Ollama connectivity.",
schema_for_empty_input(),
),
Tool::new(
"search",
"Full-text / vector / hybrid search over the knowledge base. Returns search_hit.v1 array.",
schema_for_type::<tools::search::SearchInput>(),
),
Tool::new(
"ask",
"RAG question answering over the knowledge base. Returns answer.v1 JSON. Pass session_id for multi-turn context.",
schema_for_type::<tools::ask::AskInput>(),
),
]
}
#[derive(Clone)]
pub struct KebabHandler {
state: KebabAppState,
}
impl KebabHandler {
pub fn new(state: KebabAppState) -> Self {
Self { state }
}
pub fn state(&self) -> &KebabAppState {
&self.state
}
/// Spawn a tool handler on the blocking pool. Used by tools that
/// transitively touch reqwest::blocking::Client (search, ask) — calling
/// from the async dispatch directly panics inside the runtime.
async fn spawn_tool<I, F>(
&self,
args: serde_json::Map<String, serde_json::Value>,
handle: F,
) -> Result<CallToolResult, ErrorData>
where
I: serde::de::DeserializeOwned + Send + 'static,
F: FnOnce(KebabAppState, I) -> CallToolResult + Send + 'static,
{
let input: I = match serde_json::from_value(serde_json::Value::Object(args)) {
Ok(i) => i,
Err(e) => return Ok(error::to_tool_error(&anyhow::Error::from(e))),
};
let state = self.state.clone();
tokio::task::spawn_blocking(move || handle(state, input))
.await
.map_err(|e| ErrorData::internal_error(e.to_string(), None))
}
}
impl ServerHandler for KebabHandler {
fn get_info(&self) -> ServerInfo {
ServerInfo::new(ServerCapabilities::builder().enable_tools().build())
.with_server_info(Implementation::new("kebab", env!("CARGO_PKG_VERSION")))
}
async fn list_tools(
&self,
_request: Option<rmcp::model::PaginatedRequestParams>,
_context: RequestContext<RoleServer>,
) -> Result<ListToolsResult, ErrorData> {
Ok(ListToolsResult::with_all_items(build_tools_vec()))
}
async fn call_tool(
&self,
request: CallToolRequestParams,
_context: RequestContext<RoleServer>,
) -> Result<CallToolResult, ErrorData> {
match request.name.as_ref() {
"schema" => {
let input = tools::schema::SchemaInput::default();
Ok(tools::schema::handle(&self.state, input))
}
"doctor" => {
let input = tools::doctor::DoctorInput::default();
Ok(tools::doctor::handle(&self.state, input))
}
"search" => {
let args = request.arguments.unwrap_or_default();
self.spawn_tool(args, |state, input| {
tools::search::handle(&state, input)
})
.await
}
"ask" => {
let args = request.arguments.unwrap_or_default();
self.spawn_tool(args, |state, input| {
tools::ask::handle(&state, input)
})
.await
}
_other => Err(ErrorData::method_not_found::<
rmcp::model::CallToolRequestMethod,
>()),
}
}
}
/// Run the MCP server on stdio JSON-RPC. Blocks until the client closes
/// the stream (typically when the agent host exits).
///
/// `config_path` is the path passed via `--config <path>`, if any.
/// It is forwarded to `KebabAppState` so the doctor tool can honour the
/// same config file the server was started with (falls back to XDG default
/// when `None`).
pub fn serve_stdio(cfg: Config, config_path: Option<PathBuf>) -> Result<()> {
let runtime = tokio::runtime::Builder::new_multi_thread()
.enable_all()
.build()?;
runtime.block_on(serve_stdio_async(cfg, config_path))
}
async fn serve_stdio_async(cfg: Config, config_path: Option<PathBuf>) -> Result<()> {
tracing::info!("kebab-mcp: starting stdio server");
let state = KebabAppState::new(cfg, config_path);
let handler = KebabHandler::new(state);
let service = handler.serve(stdio()).await?;
service.waiting().await?;
Ok(())
}

View File

@@ -0,0 +1,26 @@
//! Long-lived server state — holds Config so per-request handlers don't
//! reload from disk. Future: cache opened SqliteStore / Lance handles
//! here so first tool call pays the cost, subsequent calls hit warm
//! state.
use std::path::PathBuf;
use std::sync::Arc;
use kebab_config::Config;
#[derive(Clone)]
pub struct KebabAppState {
pub config: Arc<Config>,
/// `--config <path>` from CLI when present, else `None` (XDG default
/// fallback applies in `doctor_with_config_path`).
pub config_path: Option<PathBuf>,
}
impl KebabAppState {
pub fn new(config: Config, config_path: Option<PathBuf>) -> Self {
Self {
config: Arc::new(config),
config_path,
}
}
}

View File

@@ -0,0 +1,68 @@
//! `ask` tool — wraps `kebab_app::ask_with_config` (single-shot) or
//! `kebab_app::ask_with_session_with_config` when `session_id` is provided.
//! Input: { query, session_id?, mode? }. Output: answer.v1 JSON.
//!
//! `Answer` (kebab-core) does NOT carry a `schema_version` field; we tag
//! it inline here, matching the pattern from `search.rs`.
use rmcp::model::CallToolResult;
use schemars::JsonSchema;
use serde::{Deserialize, Serialize};
use crate::error::{to_tool_error, to_tool_success};
use crate::state::KebabAppState;
#[derive(Debug, Deserialize, Serialize, JsonSchema)]
pub struct AskInput {
/// The user question.
pub query: String,
/// Optional session id for multi-turn RAG context.
pub session_id: Option<String>,
/// Optional retrieval mode override ("lexical" / "vector" / "hybrid"). Default "hybrid".
pub mode: Option<String>,
}
pub fn handle(state: &KebabAppState, input: AskInput) -> CallToolResult {
let mode = match input.mode.as_deref() {
Some("lexical") => kebab_core::SearchMode::Lexical,
Some("vector") => kebab_core::SearchMode::Vector,
_ => kebab_core::SearchMode::Hybrid, // default + "hybrid" + unknown
};
let opts = kebab_app::AskOpts {
k: 10,
explain: false,
mode,
temperature: None,
seed: None,
stream_sink: None,
history: Vec::new(),
conversation_id: None,
turn_index: None,
};
let cfg_clone = (*state.config).clone();
let result = match input.session_id {
Some(sid) => {
kebab_app::ask_with_session_with_config(cfg_clone, &sid, &input.query, opts)
}
None => kebab_app::ask_with_config(cfg_clone, &input.query, opts),
};
match result {
Ok(answer) => {
// `Answer` does not carry `schema_version`; tag inline (idempotent
// via entry().or_insert_with in case a future version adds it).
let mut v = match serde_json::to_value(&answer) {
Ok(v) => v,
Err(e) => return to_tool_error(&anyhow::anyhow!("answer serialize failed: {e}")),
};
if let serde_json::Value::Object(ref mut map) = v {
map.entry("schema_version".to_string())
.or_insert_with(|| serde_json::Value::String("answer.v1".to_string()));
}
match serde_json::to_string(&v) {
Ok(json) => to_tool_success(json),
Err(e) => to_tool_error(&anyhow::anyhow!(e)),
}
}
Err(e) => to_tool_error(&e),
}
}

View File

@@ -0,0 +1,28 @@
//! `doctor` tool — wraps `kebab_app::doctor_with_config_path`.
//! Input: {} (no args). Output: doctor.v1 JSON.
//!
//! `doctor_with_config_path(Option<&Path>)` re-reads config from disk so
//! the report reflects the live file state. We forward `config_path` from
//! `KebabAppState` so `--config <path>` users see results for their file;
//! callers that pass `None` fall back to the XDG default (same as the CLI
//! bare `kebab doctor`).
use rmcp::model::CallToolResult;
use schemars::JsonSchema;
use serde::{Deserialize, Serialize};
use crate::error::{to_tool_error, to_tool_success};
use crate::state::KebabAppState;
#[derive(Debug, Default, Deserialize, Serialize, JsonSchema)]
pub struct DoctorInput {}
pub fn handle(state: &KebabAppState, _input: DoctorInput) -> CallToolResult {
match kebab_app::doctor_with_config_path(state.config_path.as_deref()) {
Ok(report) => match serde_json::to_string(&report) {
Ok(json) => to_tool_success(json),
Err(e) => to_tool_error(&anyhow::anyhow!(e)),
},
Err(e) => to_tool_error(&e),
}
}

View File

@@ -0,0 +1,6 @@
//! Tool implementations — one module per tool.
pub mod schema;
pub mod doctor;
pub mod search;
pub mod ask;

View File

@@ -0,0 +1,22 @@
//! `schema` tool — wraps `kebab_app::schema_with_config`.
//! Input: {} (no args). Output: schema.v1 JSON.
use rmcp::model::CallToolResult;
use serde::{Deserialize, Serialize};
use schemars::JsonSchema;
use crate::error::{to_tool_error, to_tool_success};
use crate::state::KebabAppState;
#[derive(Debug, Default, Deserialize, Serialize, JsonSchema)]
pub struct SchemaInput {}
pub fn handle(state: &KebabAppState, _input: SchemaInput) -> CallToolResult {
match kebab_app::schema_with_config(&state.config) {
Ok(report) => match serde_json::to_string(&report) {
Ok(json) => to_tool_success(json),
Err(e) => to_tool_error(&anyhow::anyhow!(e)),
},
Err(e) => to_tool_error(&e),
}
}

View File

@@ -0,0 +1,71 @@
//! `search` tool — wraps `kebab_app::search_with_config`.
//! Input: { query, mode?, k? }. Output: search_hit.v1 array JSON.
//!
//! First tool with a non-empty `inputSchema`: `SearchInput` derives
//! `JsonSchema` and `Tool::new` uses
//! `rmcp::handler::server::common::schema_for_type::<SearchInput>()`.
use rmcp::model::CallToolResult;
use schemars::JsonSchema;
use serde::{Deserialize, Serialize};
use crate::error::{to_tool_error, to_tool_success};
use crate::state::KebabAppState;
#[derive(Debug, Deserialize, Serialize, JsonSchema)]
pub struct SearchInput {
/// User query (free text).
pub query: String,
/// Retrieval mode: "hybrid" (default), "lexical", or "vector".
#[serde(default = "default_mode")]
pub mode: String,
/// Top-K results. Defaults to 10. Clamped to 1100.
#[serde(default = "default_k")]
pub k: usize,
}
fn default_mode() -> String {
"hybrid".to_string()
}
fn default_k() -> usize {
10
}
pub fn handle(state: &KebabAppState, input: SearchInput) -> CallToolResult {
let k = input.k.clamp(1, 100);
let mode = match input.mode.as_str() {
"lexical" => kebab_core::SearchMode::Lexical,
"vector" => kebab_core::SearchMode::Vector,
_ => kebab_core::SearchMode::Hybrid,
};
let query = kebab_core::SearchQuery {
text: input.query,
mode,
k,
filters: kebab_core::SearchFilters::default(),
};
match kebab_app::search_with_config((*state.config).clone(), query) {
Ok(hits) => {
// SearchHit (kebab-core) does not carry a `schema_version` field,
// so we tag each element inline before serialising.
let tagged: Vec<serde_json::Value> = hits
.iter()
.map(|h| {
let mut v = serde_json::to_value(h).unwrap_or_default();
if let serde_json::Value::Object(ref mut map) = v {
map.insert(
"schema_version".to_string(),
serde_json::Value::String("search_hit.v1".to_string()),
);
}
v
})
.collect();
match serde_json::to_string(&serde_json::Value::Array(tagged)) {
Ok(json) => to_tool_success(json),
Err(e) => to_tool_error(&anyhow::anyhow!(e)),
}
}
Err(e) => to_tool_error(&e),
}
}

View File

@@ -0,0 +1,36 @@
//! tools/call with bad config → isError=true + error.v1 content.
use kebab_config::Config;
use kebab_mcp::{KebabAppState, KebabHandler};
use rmcp::model::RawContent;
#[tokio::test]
async fn schema_tool_emits_error_v1_when_db_missing() {
// Point at a directory that does NOT have kebab.sqlite.
let dir = tempfile::tempdir().unwrap();
let mut cfg = Config::defaults();
cfg.storage.data_dir = dir.path().to_string_lossy().into_owned();
cfg.workspace.root = dir.path().join("notes").to_string_lossy().into_owned();
cfg.models.embedding.provider = "none".to_string();
cfg.models.embedding.dimensions = 0;
// Note: NO ingest call — kebab.sqlite is absent → schema_with_config
// calls open_existing → NotIndexed → tool error.
let state = KebabAppState::new(cfg, None);
let handler = KebabHandler::new(state);
let result = kebab_mcp::tools::schema::handle(
handler.state(),
kebab_mcp::tools::schema::SchemaInput::default(),
);
assert_eq!(result.is_error, Some(true), "expected isError=true on missing DB");
let content = result.content.first().unwrap();
let text = match &content.raw {
RawContent::Text(t) => &t.text,
other => panic!("expected text content, got {other:?}"),
};
let v: serde_json::Value = serde_json::from_str(text).unwrap();
assert_eq!(v.get("schema_version").and_then(|s| s.as_str()), Some("error.v1"));
assert_eq!(v.get("code").and_then(|s| s.as_str()), Some("not_indexed"));
}

View File

@@ -0,0 +1,19 @@
//! Integration: KebabHandler::get_info returns correct kebab serverInfo.
//! Doesn't exercise full transport — that lands when we have at least
//! one tool to call (Task 4+).
use kebab_config::Config;
use kebab_mcp::{KebabAppState, KebabHandler};
use rmcp::ServerHandler;
#[tokio::test]
async fn initialize_returns_kebab_server_info() {
let cfg = Config::defaults();
let state = KebabAppState::new(cfg, None);
let handler = KebabHandler::new(state);
let info = handler.get_info();
assert_eq!(info.server_info.name, "kebab");
assert!(!info.server_info.version.is_empty());
assert!(info.capabilities.tools.is_some());
}

View File

@@ -0,0 +1,92 @@
//! `ask` tool returns answer.v1 — refusal path covered (no Ollama
//! required for refusal-on-empty-corpus case).
use kebab_config::Config;
use kebab_core::SourceScope;
use kebab_mcp::{KebabAppState, KebabHandler};
use rmcp::model::RawContent;
fn minimal_config(data_dir: &std::path::Path, workspace_root: &std::path::Path) -> Config {
let mut cfg = Config::defaults();
cfg.storage.data_dir = data_dir.to_string_lossy().into_owned();
cfg.storage.model_dir = data_dir
.join("models")
.to_string_lossy()
.into_owned();
cfg.workspace.root = workspace_root.to_string_lossy().into_owned();
cfg.workspace.exclude.clear();
cfg.models.embedding.provider = "none".to_string();
cfg.models.embedding.dimensions = 0;
cfg
}
#[tokio::test]
async fn ask_tool_returns_answer_v1_with_refusal_on_empty_kb() {
let dir = tempfile::tempdir().unwrap();
let data_dir = dir.path().join("data");
let workspace_root = dir.path().join("notes");
std::fs::create_dir_all(&data_dir).unwrap();
std::fs::create_dir_all(&workspace_root).unwrap();
let cfg = minimal_config(&data_dir, &workspace_root);
// Seed kebab.sqlite (empty corpus — no documents ingested).
let scope = SourceScope {
root: workspace_root.clone(),
include: vec![],
exclude: vec![],
};
let _ = kebab_app::ingest_with_config(cfg.clone(), scope, false).unwrap();
let state = KebabAppState::new(cfg, None);
let handler = KebabHandler::new(state);
// `ask_with_config` builds a `reqwest::blocking::Client` internally (for
// `OllamaLanguageModel`), which spins up and drops a tokio runtime — that
// panics when called from inside an async context. Run it on the blocking
// thread pool to avoid the conflict.
let state_clone = handler.state().clone();
let result = tokio::task::spawn_blocking(move || {
kebab_mcp::tools::ask::handle(
&state_clone,
kebab_mcp::tools::ask::AskInput {
query: "what is the meaning of life".to_string(),
session_id: None,
// Test env uses provider="none" — Hybrid would hard-error on embedding.
// Pass Lexical explicitly so the test stays functional.
mode: Some("lexical".to_string()),
},
)
})
.await
.unwrap();
// Empty KB → refusal (grounded:false) is normal — NOT isError.
assert!(
!result.is_error.unwrap_or(false),
"expected isError=false on refusal, got {:?}",
result
);
let content = result
.content
.first()
.expect("expected at least one content item");
let text = match &content.raw {
RawContent::Text(t) => &t.text,
other => panic!("expected text content, got {other:?}"),
};
let v: serde_json::Value = serde_json::from_str(text).unwrap();
assert_eq!(
v.get("schema_version").and_then(|s| s.as_str()),
Some("answer.v1"),
"response should carry schema_version=answer.v1"
);
assert_eq!(
v.get("grounded").and_then(|b| b.as_bool()),
Some(false),
"empty KB should produce grounded=false"
);
}

View File

@@ -0,0 +1,50 @@
//! Integration: tools/call name=doctor — returns doctor.v1.
use kebab_config::Config;
use kebab_mcp::{KebabAppState, KebabHandler};
use rmcp::model::RawContent;
#[tokio::test]
async fn doctor_tool_returns_doctor_v1_json() {
let dir = tempfile::tempdir().unwrap();
let mut cfg = Config::defaults();
cfg.storage.data_dir = dir.path().join("data").to_string_lossy().into_owned();
cfg.workspace.root = dir.path().join("notes").to_string_lossy().into_owned();
cfg.models.embedding.provider = "none".to_string();
cfg.models.embedding.dimensions = 0;
std::fs::create_dir_all(&cfg.workspace.root).unwrap();
// Pass None for config_path — doctor falls back to XDG default probe
// (path won't exist in the tempdir, which is fine; doctor reports it
// as missing / error rather than panicking).
let state = KebabAppState::new(cfg, None);
let handler = KebabHandler::new(state);
let result = kebab_mcp::tools::doctor::handle(
handler.state(),
kebab_mcp::tools::doctor::DoctorInput::default(),
);
let content = result
.content
.first()
.expect("expected at least one content item");
let text = match &content.raw {
RawContent::Text(t) => &t.text,
other => panic!("expected text content, got {other:?}"),
};
let v: serde_json::Value = serde_json::from_str(text).unwrap();
assert_eq!(
v.get("schema_version").and_then(|s| s.as_str()),
Some("doctor.v1"),
"unexpected schema_version in: {v}"
);
// `ok` boolean must be present (value may be false in CI where Ollama
// is not reachable — that's expected and acceptable).
assert!(
v.get("ok").and_then(|b| b.as_bool()).is_some(),
"`ok` field missing in doctor.v1 response: {v}"
);
}

View File

@@ -0,0 +1,75 @@
//! Integration: tools/call name=schema — verify response is schema.v1.
use std::fs;
use kebab_config::Config;
use kebab_core::SourceScope;
use kebab_mcp::{KebabAppState, KebabHandler};
use rmcp::model::RawContent;
fn minimal_config(data_dir: &std::path::Path, workspace_root: &std::path::Path) -> Config {
let mut cfg = Config::defaults();
cfg.storage.data_dir = data_dir.to_string_lossy().into_owned();
cfg.storage.model_dir = data_dir
.join("models")
.to_string_lossy()
.into_owned();
cfg.workspace.root = workspace_root.to_string_lossy().into_owned();
cfg.workspace.exclude.clear();
cfg.models.embedding.provider = "none".to_string();
cfg.models.embedding.dimensions = 0;
cfg
}
#[tokio::test]
async fn schema_tool_returns_schema_v1_json() {
let dir = tempfile::tempdir().unwrap();
let data_dir = dir.path().join("data");
let workspace_root = dir.path().join("notes");
fs::create_dir_all(&data_dir).unwrap();
fs::create_dir_all(&workspace_root).unwrap();
let config = minimal_config(&data_dir, &workspace_root);
// Seed kebab.sqlite via 0-file ingest so open_existing succeeds later.
let scope = SourceScope {
root: workspace_root.clone(),
include: vec![],
exclude: vec![],
};
let _ = kebab_app::ingest_with_config(config.clone(), scope, false).unwrap();
let state = KebabAppState::new(config, None);
let handler = KebabHandler::new(state);
let result = kebab_mcp::tools::schema::handle(
handler.state(),
kebab_mcp::tools::schema::SchemaInput::default(),
);
assert!(
!result.is_error.unwrap_or(false),
"expected isError=false on healthy schema, got {:?}",
result
);
let content = result.content.first().expect("expected at least one content item");
// Content = Annotated<RawContent>; deref to get the inner RawContent.
let text = match &content.raw {
RawContent::Text(t) => &t.text,
other => panic!("expected text content, got {other:?}"),
};
let v: serde_json::Value = serde_json::from_str(text).unwrap();
assert_eq!(
v.get("schema_version").and_then(|s| s.as_str()),
Some("schema.v1"),
"unexpected schema_version in: {v}"
);
assert_eq!(
v.get("capabilities").and_then(|c| c.get("mcp_server")).and_then(|b| b.as_bool()),
Some(true),
"mcp_server capability flag should be true after fb-30",
);
}

View File

@@ -0,0 +1,90 @@
//! Integration: tools/call name=search — verify response is search_hit.v1 array.
use std::fs;
use kebab_config::Config;
use kebab_core::SourceScope;
use kebab_mcp::{KebabAppState, KebabHandler};
use rmcp::model::RawContent;
fn minimal_config(data_dir: &std::path::Path, workspace_root: &std::path::Path) -> Config {
let mut cfg = Config::defaults();
cfg.storage.data_dir = data_dir.to_string_lossy().into_owned();
cfg.storage.model_dir = data_dir
.join("models")
.to_string_lossy()
.into_owned();
cfg.workspace.root = workspace_root.to_string_lossy().into_owned();
cfg.workspace.exclude.clear();
cfg.models.embedding.provider = "none".to_string();
cfg.models.embedding.dimensions = 0;
cfg
}
#[tokio::test]
async fn search_tool_returns_search_hits_array() {
let dir = tempfile::tempdir().unwrap();
let data_dir = dir.path().join("data");
let workspace_root = dir.path().join("notes");
fs::create_dir_all(&data_dir).unwrap();
fs::create_dir_all(&workspace_root).unwrap();
let config = minimal_config(&data_dir, &workspace_root);
// Write a markdown document containing the query term.
fs::write(
workspace_root.join("a.md"),
"# Alpha\n\nThis document mentions kebab and bread.",
)
.unwrap();
// Seed kebab.sqlite via ingest so search has indexed content.
let scope = SourceScope {
root: workspace_root.clone(),
include: vec![],
exclude: vec![],
};
let _ = kebab_app::ingest_with_config(config.clone(), scope, false).unwrap();
let state = KebabAppState::new(config, None);
let handler = KebabHandler::new(state);
let result = kebab_mcp::tools::search::handle(
handler.state(),
kebab_mcp::tools::search::SearchInput {
query: "kebab".to_string(),
mode: "lexical".to_string(),
k: 5,
},
);
assert!(
!result.is_error.unwrap_or(false),
"expected isError=false, got {:?}",
result
);
let content = result
.content
.first()
.expect("expected at least one content item");
let text = match &content.raw {
RawContent::Text(t) => &t.text,
other => panic!("expected text content, got {other:?}"),
};
let v: serde_json::Value = serde_json::from_str(text).unwrap();
let arr = v.as_array().expect("search returns a JSON array");
assert!(
!arr.is_empty(),
"expected at least one hit for 'kebab' in 'a.md'"
);
assert_eq!(
arr[0]
.get("schema_version")
.and_then(|s| s.as_str()),
Some("search_hit.v1"),
"first hit should carry schema_version=search_hit.v1"
);
}

View File

@@ -0,0 +1,68 @@
//! Integration: `build_tools_vec` returns 4 tools with correct names and
//! inputSchema. Uses the extracted `pub fn build_tools_vec()` helper — no
//! transport or RequestContext needed.
use kebab_mcp::build_tools_vec;
#[test]
fn tools_list_returns_four_tools() {
let tools = build_tools_vec();
assert_eq!(tools.len(), 4, "expected exactly 4 tools, got {}", tools.len());
let names: Vec<&str> = tools.iter().map(|t| t.name.as_ref()).collect();
assert!(names.contains(&"schema"), "missing 'schema' tool");
assert!(names.contains(&"doctor"), "missing 'doctor' tool");
assert!(names.contains(&"search"), "missing 'search' tool");
assert!(names.contains(&"ask"), "missing 'ask' tool");
}
#[test]
fn search_tool_input_schema_has_required_query() {
let tools = build_tools_vec();
let search = tools
.iter()
.find(|t| t.name.as_ref() == "search")
.expect("search tool must be present");
// input_schema is Arc<JsonObject> (serde_json::Map<String, Value>).
let schema_val = serde_json::Value::Object(search.input_schema.as_ref().clone());
let required = schema_val
.get("required")
.and_then(|v| v.as_array())
.expect("search inputSchema must have a 'required' array");
assert!(
required.iter().any(|v| v.as_str() == Some("query")),
"search inputSchema 'required' must contain 'query', got: {required:?}"
);
}
#[test]
fn schema_and_doctor_tools_accept_empty_input() {
let tools = build_tools_vec();
for name in &["schema", "doctor"] {
let tool = tools
.iter()
.find(|t| t.name.as_ref() == *name)
.unwrap_or_else(|| panic!("{name} tool must be present"));
let schema_val = serde_json::Value::Object(tool.input_schema.as_ref().clone());
// An empty-input schema has type "object" and no required fields
// (or no 'required' key at all).
let ty = schema_val.get("type").and_then(|v| v.as_str());
assert_eq!(
ty,
Some("object"),
"{name} inputSchema 'type' must be 'object', got {ty:?}"
);
if let Some(required) = schema_val.get("required").and_then(|v| v.as_array()) {
assert!(
required.is_empty(),
"{name} inputSchema 'required' must be empty, got: {required:?}"
);
}
}
}

View File

@@ -40,6 +40,7 @@ flowchart TB
subgraph UI ["UI binary"]
cli["kebab-cli"]
tui["kebab-tui"]
mcp["kebab-mcp<br/>(P9-FB-30)"]
desktop["kebab-desktop<br/>(P9-5)"]
end
app["kebab-app<br/>(facade)"]
@@ -71,6 +72,7 @@ flowchart TB
cli --> app
tui --> app
mcp --> app
desktop --> app
app --> srcfs
@@ -168,6 +170,7 @@ kebab/
│ ├── kebab-parse-pdf/ # lopdf per-page text extractor (P7-1)
│ ├── kebab-app/ # facade (P0 시그니처 + P3-5/P6-4/P7-3 본체)
│ ├── kebab-tui/ # Ratatui shell + Library 패널 (P9-1)
│ ├── kebab-mcp/ # stdio MCP server — tools: schema, doctor, search, ask (P9-FB-30)
│ └── kebab-cli/ # binary (P0 → 핫픽스로 --config flag wiring 강화)
├── migrations/ # SQLite refinery V001/V002/V003
└── fixtures/ # 테스트 fixture 트리

View File

@@ -1443,6 +1443,17 @@ HOTFIXES 의 `2026-05-07 — p9-fb-27` 항목이 details shape 의
interim deviation (IoFailure / OpTimeout 신규 typed signal 도입 전까지의
transitional 형태) 의 source of truth.
### 10.2 MCP server transport (fb-30)
`kebab mcp` 가 stdio JSON-RPC server. Rust SDK = `rmcp 1.6`. Tool surface
v1: `search` / `ask` / `schema` / `doctor` (4 read-only). Resources /
Prompts / Sampling 미선언. Output 은 wire schema v1 JSON 을 MCP `text`
content block 으로 직렬화. Tool dispatch 실패는 `isError: true` + error.v1
content; refusal / no-hit / unhealthy 는 정상 응답 (semantic flag 으로
agent 가 분기). HTTP-SSE transport 는 fb-29 deferral 따라 P+. classify
모듈은 `kebab-app::error_wire` 에 single source — kebab-cli + kebab-mcp
공유.
---
## 11. 동결 범위 / 변경 정책

View File

@@ -71,6 +71,25 @@ Returns a `schema.v1` object with: `wire.schemas` (supported wire ids), `capabil
If a call fails or returns suspicious output, run `kebab doctor` first — it surfaces config-load / data-dir / Ollama-reachability problems in one line each. Don't silently retry on errors; report the doctor output.
## MCP server (recommended over CLI subprocess wrapping)
Since v0.4.0, `kebab` exposes an MCP (Model Context Protocol) stdio server. Configure once in `~/.claude/mcp.json`:
```json
{
"mcpServers": {
"kebab": {
"command": "kebab",
"args": ["mcp"]
}
}
}
```
Claude Code spawns `kebab mcp` at session start; the process stays alive across all tool calls so SQLite / Lance / fastembed are hot after the first call. 4 tools available: `search` / `ask` / `schema` / `doctor`. Same wire shapes as the CLI `--json` mode — see `Two surfaces, pick the right one` above for the same guidance.
If your host doesn't support MCP, the CLI subprocess pattern (`kebab search --json` / `kebab ask --json`) above continues to work.
## Workflow recipes
**Recipe A — user asks an internal-context question, you want grounded answer:**

View File

@@ -14,6 +14,46 @@ historical contract that was implemented; this file accumulates the
deltas so phase 5+ readers can find the live behavior without diffing
git history.
## 2026-05-07 — p9-fb-30 (post-dogfooding): MCP server (stdio) — agent integration MVP
**Source feedback**: 사용자 도그푸딩 2026-05-06 — Claude Code 같은 AI agent 가 kebab CLI 를 사용하는 것이 궁극 목표. 현재 surface 는 Claude Code 전용 skill (subprocess wrapper) 만 — host 무관 표준 통신 없음. fb-29 HTTP daemon 은 single-user local-first 환경 대비 비대로 deferred (2026-05-07), fb-30 stdio MCP 가 동일 사용자 가치 (agent integration + session 동안 hot cache) 를 daemon 복잡도 없이 제공.
**Live binding 변경**:
- 신규 subcommand `kebab mcp` — stdio JSON-RPC server, `--config <path>` honor.
- 신규 crate `kebab-mcp` (lib only) — `serve_stdio(Config, Option<PathBuf>)` entry. UI crate 카테고리 (kebab-cli + kebab-tui + kebab-mcp 가 facade 룰 동일 적용 — `kebab-app` facade 만 import).
- Tool surface v1 (read-only 4): `search` (lexical/vector/hybrid 검색, default Hybrid), `ask` (RAG 답변, default mode Hybrid, optional `session_id` for multi-turn + optional `mode` override), `schema` (introspection), `doctor` (health check). `ingest_*` / `fetch` / `list_docs` / `inspect_chunk` 는 fb-31 / fb-35 / 후속 task 머지 시 추가.
- Resources / Prompts / Sampling — 모두 미선언 (tools-only v1).
- Output: 모든 tool 이 wire schema v1 JSON 을 MCP `text` content block 으로 직렬화. CLI `--json` 모드와 동일 wire — single source.
- Error mapping: tool dispatch `Err(e)``isError: true` + error.v1 content. Refusal (`grounded: false`) / no-hit (empty array) / unhealthy (`ok: false`) 는 모두 정상 응답 — agent 가 wire payload semantic flag 으로 분기.
- `kebab-app::error_wire` 신규 — fb-27 의 `kebab-cli::error_classify` 코드 그대로 promotion (struct + classify + classify_llm + 7 unit test). kebab-cli + kebab-mcp 둘 다 동일 모듈 사용. reqwest dev-dep 도 함께 이동. 부수 변경: `ErrorV1``schema_version: String` 필드 추가 — kebab-mcp 의 직접 serialize 경로에서도 wire 정합 (kebab-cli 의 `wire_error_v1``tag_object` 는 idempotent 로 작동, 동작 무영향).
- `kebab-app::Capabilities::mcp_server`: `false``true`. `schema_report` 통합 테스트 + `cli_schema` 통합 테스트 assertion 갱신.
- Initialize handshake: `protocolVersion = "2025-03-26"` (rmcp 1.6 default), `capabilities.tools = { listChanged: false }`, `serverInfo = { name: "kebab", version: <CARGO_PKG_VERSION> }`.
- `KebabAppState``(Config, Option<PathBuf>)` carry — `kebab_app::doctor_with_config_path``Option<&Path>` 만 받기 때문 (`doctor_with_config(&Config)` 미존재). path 없으면 `None` (XDG default 동작).
- `tokio::task::spawn_blocking` wrap on `call_tool` arms for `ask` + `search``OllamaLanguageModel``reqwest::blocking::Client::build()` 가 내부적으로 tokio runtime create+drop 하므로 async 안에서 panic. spawn_blocking 으로 우회. schema / doctor 는 cheap reads 라 wrap 불필요.
- `tools/list` 의 list construction 을 `pub fn build_tools_vec()` 로 추출 — rmcp 1.6 가 in-memory test transport 미노출이라 spawn 없이 unit-level 검증 위함.
**Spec contract impact**: design §10 에 §10.2 MCP transport 절 추가.
**Tests added**: kebab-mcp integration (5: tools_call_search / tools_call_ask / tools_call_schema / tools_call_doctor / tools_list / error_mapping + initialize), kebab-cli integration (1: cli_mcp_smoke spawn + initialize + tools/list round-trip). 약 8 신규 테스트.
**Known limitation (deferred)**:
- HTTP-SSE transport — fb-29 P+ deferral 따라 stdio 단일. browser agent / remote 시나리오 등장 시 재개.
- Resources (`kebab://chunk/<id>` URI) — fb-35 verbatim fetch 와 함께 v2.
- Prompts — RAG 자체 prompt template 내장으로 사용자 가치 약함, defer.
- Streaming `ask` — fb-33 streaming ask 와 함께.
- `ingest_*` / `fetch` / `list_docs` / `inspect_chunk` tools — 후속 task 별로 추가.
- Server-scope state caching — 현재 매 tool call 마다 store open. 첫 call 시 `KebabAppState``OnceLock<SqliteStore>` 도입 검토 (post-merge 후속 PR).
- rmcp SDK API 호환성 — 1.6 채택, 미래 major bump 시 별 task.
- Manual `tools/list` + `tools/call` dispatch 채택 — rmcp 1.6 의 `#[tool_router]` 매크로보다 명시적, 디버깅 쉬움. 하지만 새 tool 추가 시 두 곳 (list_tools 의 vec + call_tool 의 match) 동시 갱신 필요. 후속 task 가 5개 이상 tool 추가하면 매크로 도입 재검토.
- `AskOpts``Default` 미도입 — kebab-cli + kebab-tui + kebab-mcp 의 모든 호출 site 가 9 field 를 명시적으로 초기화. 새 field 추가 시 모든 site 동시 갱신 필요. `impl Default for AskOpts` 또는 builder 패턴 도입은 별 PR.
**Amends**:
- design §10 (MCP transport subsection 추가).
- spec `tasks/p9/p9-fb-30-mcp-server.md` (status `open``completed`).
- spec stub 의 `transport: stdio default + http (fb-29 daemon) 위에 SSE 옵션` → 실제 채택 stdio 단일 (fb-29 deferral 결과, 2026-05-07 commit `2e8de14` 의 spec 갱신과 일관).
## 2026-05-07 — p9-fb-27 (post-dogfooding): introspection (`kebab schema`) + structured error wire
**Source feedback**: 사용자 도그푸딩 2026-05-06 — agent 가 kebab 인스턴스의 wire 버전 / 기능 / 모델 / 인덱스 통계 introspect 못 함; error 가 stderr text 라 substring 분기 필요.

View File

@@ -3,7 +3,7 @@ phase: P9
component: integrations + new crate (kebab-mcp)
task_id: p9-fb-30
title: "MCP server — agent host 무관 protocol surface"
status: open
status: completed
target_version: 0.3.0
depends_on: [p9-fb-27]
unblocks: []
@@ -14,7 +14,7 @@ source_feedback: 사용자 도그푸딩 2026-05-06 — Claude Code 같은 AI age
# p9-fb-30 — MCP server
> **백로그 only — 미구현.** 본 spec 은 도그푸딩 피드백 skeleton. 구현 착수 전 [superpowers:brainstorming](../../docs/superpowers/) 으로 설계 단계 선행 필요. transport 선택 (stdio / socket) / tool surface 범위 / authentication / resources vs tools 매핑 brainstorm 후 확정.
> **구현 완료.** 본 spec 은 구현 시점의 frozen 상태. post-merge deviation (특히 `error.v1` 에 schema_version 필드 추가, ask/search spawn_blocking, manual dispatch 채택) 은 [HOTFIXES.md](../HOTFIXES.md) 의 `2026-05-07 — p9-fb-30` 항목 참조 — live source of truth.
## 증상 / 동기