Skip to main content

stdio MCP server for the FCoP Python library: exposes the fcop package's Project, task, report, and issue APIs to MCP clients (Cursor, Claude Desktop, …). Optional bridge, not a second protocol; depends on fcop 1.x, FastMCP, websockets.

Project description

fcop-mcp

MCP (stdio) server — the optional IDE bridge for the same FCoP stack. It wraps the official fcop library; it is not a second “FCoP product” and does not replace the protocol text.

  • What FCoP is (protocol only, product-agnostic): docs/getting-started.en.md (中文 getting-started.md)
  • Pure Python lib / pip install fcop: filesystem + Project API, PyYAML only — PyPI fcop (see that package’s description and Documentation).
  • This package (fcop-mcp): pip install fcop-mcp — stdio tools/resources for clients; same repo, folder mcp/.
  • Source home: joinwell52-AI/FCoP

Upgrading from 0.6.x / 0.7.x? See docs/upgrade-fcop-mcp.md — run pip install -U fcop fcop-mcp in the MCP venv, restart the IDE, then run the MCP tool redeploy_rules() once to refresh on-disk protocol rule files. v1.1 is the latest stable release: adds Task.risk_level, Review.decision=needs_human, Review.human_approval, Agent.layer governance contracts, and Skill.tools[] risk metadata. v1.0 introduced the fcop/ workspace path (previously docs/agents/); see docs/MIGRATION-1.0.md for details.

What can the server actually do? The 30 MCP tools and 14 read-only resources are indexed (with grouping, when-to-call, and parameter cheatsheet) in docs/mcp-tools.md. Authoritative descriptions remain in source docstrings (mcp/src/fcop_mcp/server.py). v1.1.0 adds 4 new tools: write_review, list_reviews, read_review, mark_human_approved (human-in-the-loop approval flow), and extends write_task with a risk_level parameter — high-risk tasks automatically trigger a needs_human review. 0.6.4 adds 2 new resources (fcop://prompt/install zh + en) and gives every init_* tool a force parameter for clean ADMIN-driven team switches. 0.6.5 wires the Rule 0.a.1 hard constraint (task → do → report → archive four-step cycle) into the tool layer: new_workspace prepends a soft Rule 0.a.1 reminder when no open TASK-*.md mentions the slug, and fcop_report (initialised branch) ends with the explicit four-step template — both bilingual, both additive (no signature changes). See docs/releases/0.6.5.md.

0.6.3 ships ADR-0006 — host-neutral protocol-rule distribution. New tool fcop_report is now the canonical session/init report (its header carries a [Versions] block that flags drift between the wheel-bundled rules and the project-local .cursor/rules/ copy). New ADMIN-only tool redeploy_rules writes the four protocol-rule targets — .cursor/rules/fcop-rules.mdc, .cursor/rules/fcop-protocol.mdc, AGENTS.md, CLAUDE.md — so Cursor, Claude Code CLI, and Codex CLI all see the same rules. Legacy unbound_report stays as a deprecated alias of fcop_report (emits DeprecationWarning, removed in 0.7.0). See docs/releases/0.6.3.md for the full migration story.

The fcop on PyPI must be the FCoP library (summary mentions File-based Coordination Protocol, pyyaml, no fastmcp inside fcop). If pip show fcop says MCP toolbox or from fcop import Issue fails, you have a wrong distribution — fix with a clean venv and reinstall (see Verify below).


TL;DR — Have an agent install fcop-mcp for you

Don't want to read the rest of this page? Open a fresh chat with any shell-capable AI agent (Cursor with a new tab, Claude Code CLI, Codex CLI…) and paste this prompt verbatim. The agent will detect your OS, install uv, edit your ~/.cursor/mcp.json (preserving every other server you already have), and tell you when to restart.

Install fcop-mcp into Cursor for me — you run the commands end to end.

1. Detect my OS first: in the terminal, run
   `uname -s 2>$null; echo $env:OS` to see whether this is Windows
   or macOS / Linux.

2. Install uv (if not already present). One-liner:
   - Windows PowerShell:
     powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"
   - macOS / Linux:
     curl -LsSf https://astral.sh/uv/install.sh | sh
   Then run `uvx --version` to confirm.

3. Add an fcop entry to the global mcp.json. **Preserve** existing
   mcpServers — do not overwrite them.
   - Windows path: %USERPROFILE%\.cursor\mcp.json
   - macOS / Linux path: ~/.cursor/mcp.json
   - Add this snippet inside the mcpServers object:
     "fcop": {
       "command": "uvx",
       "args": ["fcop-mcp"]
     }

4. Print the final mcp.json contents back to me.

5. Remind me to restart Cursor; on first launch fcop-mcp will pull
   dependencies, **wait 30 seconds to 1 minute**, do not close or
   reconnect early.

Report back after each step before moving on. **Do not** auto-init
a project after install — initialization is my (ADMIN's) choice; I
will pick solo / dev-team / custom myself.

中文版本 / Chinese version: see agent-install-prompt.zh.md in the repo, or after install read the MCP resource fcop://prompt/install.

Why the "do not auto-init" line? Initialisation is ADMIN's three-way choice (solo / preset team / custom). 0.6.3 had agents defaulting to init_project(team="dev-team"), which silently overwrote ADMIN's intended Solo flow. 0.6.4 makes the prompt and the fcop_report Phase-1 message say the choice out loud and forbid the agent from picking on ADMIN's behalf.


One-page install (what we recommend for customers)

Goal: a dedicated Python environment for MCP only, so no other project’s .pth or wrong fcop shadows the real library.

A. Recommended: dedicated venv + python -m fcop_mcp

  1. Python 3.10+ on PATH (3.10–3.13 tested in CI; avoid very new 3.14 until CI covers it).
  2. Create a venv (paths are examples — adjust if you like):

Windows (PowerShell)

$v = "$env:USERPROFILE\.cursor\fcop_mcp_venv"
py -3.10 -m venv $v
& "$v\Scripts\pip.exe" install -U pip
& "$v\Scripts\pip.exe" install -U "fcop" "fcop-mcp"

macOS / Linux

VENV="$HOME/.cursor/fcop_mcp_venv"
python3 -m venv "$VENV"
"$VENV/bin/pip" install -U pip
"$VENV/bin/pip" install -U "fcop" "fcop-mcp"
  1. Cursor user config — file:
  • Windows: %USERPROFILE%\.cursor\mcp.json
  • macOS / Linux: ~/.cursor/mcp.json

Add or merge (use the real python path from step 2):

{
  "mcpServers": {
    "fcop": {
      "command": "C:\\Users\\YOUR_USER\\.cursor\\fcop_mcp_venv\\Scripts\\python.exe",
      "args": ["-m", "fcop_mcp"]
    }
  }
}

On macOS, command is like /Users/YOUR_USER/.cursor/fcop_mcp_venv/bin/python.

  1. Fully restart Cursor (or Developer: Reload Window), then open MCP and confirm fcop is connected.

Why this path? uvx (below) is convenient but first run can take a long time to download dependencies; some MCP hosts time out. A fixed venv avoids that and avoids name conflicts with other editable installs of fcop on the same machine.


B. Alternative: uvx fcop-mcp (quickest to try, slower cold start)

{
  "mcpServers": {
    "fcop": {
      "command": "uvx",
      "args": ["fcop-mcp"]
    }
  }
}

Install uv first. First connection may download many wheels — wait for it; don’t spam reconnect. If you see Aborted or timeouts, use A above.


Verify (2 commands)

In the same venv you use for MCP:

python -c "from fcop import Issue, Project; print('fcop OK', Project)"
python -c "from fcop_mcp.server import mcp; print('fcop-mcp OK')"

If the first line fails, fcop is not the FCoP library — uninstall and reinstall in a clean venv (fcop / fcop-mcp from PyPI, same MAJOR in lockstep with the current release; e.g. fcop 1.0.x + fcop-mcp 1.0.x).


Claude Desktop

  • macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
  • Windows: %APPDATA%\Claude\claude_desktop_config.json

Use the same command / args as Cursor (either A with your venv python, or B with uvx).


Upgrading from uvx / args: ["fcop"] (0.5.x)

"fcop": { "command": "uvx", "args": ["fcop-mcp"] }

The mcpServers key name can stay "fcop". Full guide:
docs/MIGRATION-0.6.md


Where the server looks for the project

Resolution order (see ADR-0003):

  1. Last set_project_dir in this MCP session
  2. FCOP_PROJECT_DIR
  3. Legacy 0.5.x env var CODEFLOW_PROJECT_DIR (still recognized with a deprecation warning — use FCOP_PROJECT_DIR)
  4. Walk up for docs/agents/fcop.json / fcop-rules.mdc / docs/agents/tasks/
  5. Current working directory

To pin a folder in config:

"env": { "FCOP_PROJECT_DIR": "D:/path/to/your/repo" }

Stability (1.x)

Tool and resource shapes are additive-only within a single MINOR release line (stability charter, ADR-0003). Configs that work on fcop-mcp 1.0.0+ will keep working on later 1.x patch and minor releases. Breaking changes are allowed only at MAJOR boundaries; patch releases never break existing tool calls.

For projects still on fcop-mcp 0.7.x, see docs/upgrade-fcop-mcp.md and docs/releases/1.0.0.md for the 0.7 → 1.0 migration narrative. For v1.0 → v1.1 (additive, no breaking changes), see docs/MIGRATION-1.1.md and docs/releases/1.1.0.md.


License

MIT — see LICENSE.

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distributions

No source distribution files available for this release.See tutorial on generating distribution archives.

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

fcop_mcp-1.1.1-py3-none-any.whl (40.5 kB view details)

Uploaded Python 3

File details

Details for the file fcop_mcp-1.1.1-py3-none-any.whl.

File metadata

  • Download URL: fcop_mcp-1.1.1-py3-none-any.whl
  • Upload date:
  • Size: 40.5 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.12.9

File hashes

Hashes for fcop_mcp-1.1.1-py3-none-any.whl
Algorithm Hash digest
SHA256 0f17d37afda2d9a7eee9b9df40f63f98e49b7712ad84d3e0082420e8bdebd803
MD5 0e292b1124d8c23c8aa375720c6c8111
BLAKE2b-256 5d1173dc3afef2738be386c5b1cc774832166a31f7960d2c37e85c265c9d43c1

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page