pitroom 0.13.0 → 0.14.0

package/.claude-plugin/plugin.json

@@ -1,7 +1,7 @@
 {
   "name": "pitroom",
   "description": "A free pit crew for your expensive coding agent: the superpowers development workflow, run by cheap workers. Delegate research, fixes, tests, reviews and whole plans from Claude Code, Codex or any agent to OpenCode, Codex or Claude Code workers, with receipts that prove the savings.",
-  "version": "0.13.0",
+  "version": "0.14.0",
   "author": {
     "name": "TahaAtas"
   },

package/.codex-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "pitroom",
-  "version": "0.13.0",
+  "version": "0.14.0",
   "description": "A free pit crew for your expensive coding agent: delegate research, fixes, tests, reviews and parallel investigations to cheap worker agents running your own models, with verified answers, exact diffs and cost receipts.",
   "author": {
     "name": "TahaAtas"

package/CHANGELOG.md

@@ -1,5 +1,15 @@
 # Changelog
 
+## 0.14.0
+
+### MCP server
+- `pitroom mcp` serves Pitroom as a [Model Context Protocol](https://modelcontextprotocol.io) server on stdio, so any agent that speaks MCP (Cursor, Claude Desktop, Claude Code, Codex, Gemini CLI, …) can use it as tools instead of shell commands and skills: `pitroom_run` (read, isolate or write), `pitroom_wait`, `pitroom_status`, `pitroom_show`, `pitroom_review`, `pitroom_audit`, `pitroom_apply`, `pitroom_discard` and `pitroom_stop`, each with a JSON schema and hints (`apply` is marked destructive). Set it up with `claude mcp add pitroom -- pitroom mcp`, `codex mcp add pitroom -- pitroom mcp`, `gemini mcp add pitroom pitroom mcp`, or an `mcpServers` entry for Cursor and Claude Desktop (see the README).
+- Each tool runs the matching `pitroom` command, so every rule of the CLI applies unchanged (permission profiles, git guard, isolation, read snapshots, cooldowns, audits) and the tools cannot drift from the CLI. A run is waited for up to `waitSeconds` (default 50, at most 540), then comes back as "still running" with its id for `pitroom_wait`. No new dependency: a small JSON-RPC implementation (protocol versions 2025-06-18, 2025-03-26 and 2024-11-05, batches, the standard error codes); only JSON reaches stdout.
+- Tested against the protocol with a real server process, and with a real client: Gemini CLI connected, called `pitroom_run` for a question, got the worker's answer with its references verified and collected an audit. Limits: stdio only, no MCP resources, prompts or progress notifications yet (the agent polls with `pitroom_wait`).
+
+### Fixes
+- `pitroom stop` on a run whose process is still starting: the process has no signal handler yet and died on the signal, and the run showed "failed: worker process exited unexpectedly". `stop` now says first that a stop was asked for, and such a run is "stopped". This was the cause of a flaky test (`stop -g …`).
+
 ## 0.13.0
 
 ### Read snapshots

package/README.md

@@ -267,7 +267,25 @@ gemini extensions install https://github.com/ATASTECH/pitroom
 npm i -g pitroom
 ```
 
-**Any other agent (Cursor, a plain shell) or just the CLI**
+**Any agent that speaks MCP (Cursor, Claude Desktop, and the three above too)**
+
+Pitroom is also an [MCP](https://modelcontextprotocol.io) server on stdio: nine tools (`pitroom_run`, `pitroom_wait`, `pitroom_status`, `pitroom_show`, `pitroom_review`, `pitroom_audit`, `pitroom_apply`, `pitroom_discard`, `pitroom_stop`) instead of shell commands and skills. After `npm i -g pitroom`:
+
+```bash
+claude mcp add pitroom -- pitroom mcp
+codex mcp add pitroom -- pitroom mcp
+gemini mcp add pitroom pitroom mcp
+```
+
+or, for Cursor (`~/.cursor/mcp.json`), Claude Desktop (`claude_desktop_config.json`) and others:
+
+```json
+{ "mcpServers": { "pitroom": { "command": "pitroom", "args": ["mcp"] } } }
+```
+
+Each tool runs the matching `pitroom` command, so every rule of the CLI applies unchanged (permission profiles, git guard, isolation, read snapshots). A run is waited for up to `waitSeconds` (default 50), then comes back as "still running" with its id for `pitroom_wait`; keep it below your client's tool timeout. The server works in the directory the client starts it in.
+
+**Any other agent (a plain shell) or just the CLI**
 
 ```bash
 npm i -g pitroom
@@ -278,6 +296,7 @@ pitroom install
 |---|---|
 | Claude Code plugin | The 14 skills, the session-start hook that introduces Pitroom, and a card after each `pitroom` command. |
 | Gemini CLI extension | The 14 skills and a short context file that introduces Pitroom (Gemini asks you to confirm the extension). Gemini CLI also works as a *worker*, which is separate: see [Workers](#workers). |
+| MCP server | The nine tools above, for any agent that can use MCP servers. No skills: the tools describe themselves, and the server tells the agent how to start. |
 | Codex plugin | The 14 skills, a session-start hook that introduces Pitroom (so it is known even when Codex drops skill descriptions because many are installed), and a card after each `pitroom` command. Codex asks you to trust the plugin's hooks once. The hooks call `pitroom`, which comes from npm (`npm i -g pitroom`). |
 | npm + `pitroom install` | The CLI, and the skills linked into `~/.agents/skills` and `~/.claude/skills`. |
 
@@ -459,6 +478,7 @@ pitroom crew [-i] [-g NAME] "task 1" "task 2" …   (or --task-file with --- sep
 pitroom run -i --plan PLAN --step N [--tier T] ["notes"]
 pitroom review [run | --range A..B [--plan PLAN]] [--tier T | -W T] [--bg]
 pitroom audit RUN [-W worker]
+pitroom mcp                  # serve Pitroom as MCP tools on stdio
 pitroom cooldown [--clear]
 pitroom plan status PLAN [--json] · pitroom plan note PLAN "Task N: …"
 pitroom status|wait|watch [run… | -g NAME]        wait: --any --brief --timeout · watch: --json|--brief --interval