About fcop
fcop-mcp adapts the FCoP Python Core for MCP clients such as Codex, Cursor, and Claude Desktop. It exposes 25 canonical tools and six read-only resources by default, while Core owns protocol validation and state management.
Key features
- 0125 canonical MCP tools
- 02Six read-only Core resources
- 03Core-managed protocol validation and state transitions
- 04Local stdio server support
- 05CLI for setup, inspection, and diagnosis
Use cases
- 01Create and manage FCoP tasks from an MCP client
- 02Inspect workspace status and validate protocol structure
- 03Diagnose installation and package-version compatibility
MCP support
fcop-mcp connects FCoP Core to MCP clients and exposes 25 tools and six read-only resources over local stdio by default. Core remains authoritative for protocol facts, validation, persistence, and recovery.
Configuration
Local server config is the same for every client. Copy it into your client’s MCP config file.
uvx fcop-mcpFCOP_PROJECT_DIRAbsolute path to your project root (optional, auto-detected if omitted).Local servers need a runtime on your machine (e.g. Node.js 18+). Config file locations differ by client; see your client’s docs.
Tools
25Saved tool list · updated 2w ago
| Parameter | Type | Required | Description |
|---|---|---|---|
| profile | any | No |
Details
From the project READMEfcop-mcp · Canonical MCP adapter for FCoP
mcp-name: io.github.joinwell52-AI/fcop
fcop-mcp exposes the official FCoP Python Core
to Codex, Cursor, Claude Desktop, and other MCP clients. It is a thin adapter
over Core, which remains the sole authority for protocol facts, validation,
state transitions, persistence, locking, and recovery.
Current version: 4.0.5. The default server exposes exactly 25 Canonical MCP Tools and 6 read-only Core Resources. The tool registry is derived from one canonical manifest. The resource registry projects versioned content owned by Core.
Install
Install the matching 4.0.5 package pair in a dedicated Python 3.10–3.13 environment:
python -m pip install "fcop==4.0.5" "fcop-mcp==4.0.5"
fcop-mcp==4.0.5 declares fcop>=4.0.5,<4.1.0. Verify both distributions:
python -c "from importlib.metadata import version; print('fcop', version('fcop')); print('fcop-mcp', version('fcop-mcp'))"
python -c "from fcop_mcp.server import mcp; print('fcop-mcp ready')"
fcop and fcop-mcp are released in lockstep within the same minor line.
Do not mix incompatible minor versions.
macOS (Intel and Apple silicon)
FCoP MCP supports both Intel and Apple silicon Macs; Rosetta is not required. With Python 3.10–3.13:
python3 -m venv ~/.local/share/fcop/venv
~/.local/share/fcop/venv/bin/python -m pip install --upgrade \
"fcop>=4.0.5,<4.1.0" \
"fcop-mcp>=4.0.5,<4.1.0"
~/.local/share/fcop/venv/bin/fcop version
~/.local/share/fcop/venv/bin/fcop doctor
~/.local/share/fcop/venv/bin/fcop tools --json
The installed CLI provides init, status, inspect, validate, tools, doctor, version, spec, and migrate. MCP additionally requires a client that supports local stdio servers. Configure the command as the absolute /Users/.../.local/share/fcop/venv/bin/python path, pass ["-m", "fcop_mcp"], and set FCOP_PROJECT_DIR to the actual project root.
For an isolated launcher:
uvx --from fcop-mcp==4.0.5 --with fcop==4.0.5 fcop-mcp --root /absolute/path/to/repository
Canonical 4.0.5 surface
The workspace tools include init_workspace, inspect_workspace, and
validate_workspace. create_task is the canonical TASK creation entry.
Profiles are explicit extensions. Toolkit commands, compatibility shims,
packaging operations, Host configuration, Runtime services, and application
scaffolding are outside the default MCP registry.
Team models, Solo mode, ME or other fixed roles, seats, and Host sessions are
not Base Protocol prerequisites. The default server neither requires nor
creates them. init_workspace does not generate AGENTS.md, CLAUDE.md,
Cursor rules, role templates, or session state.
The prior inventory is the historical 49-tool implementation surface. It
is not the current MCP contract. Every historical tool has one disposition in
the 4.0.5 migration guide.
For example, write_task is an opt-in deprecated Python compatibility shim,
and fcop_audit is a Toolkit CLI command. Neither appears in default
tools/list.
See the generated MCP tools reference for all 25 names, descriptions, and input schemas.
Configure an MCP client
A fixed virtual environment gives predictable startup and avoids importing a different editable package from another project.
{
"mcpServers": {
"fcop": {
"command": "/absolute/path/to/venv/bin/python",
"args": ["-m", "fcop_mcp", "--root", "/absolute/path/to/repository"]
}
}
}
On Windows, command is the virtual environment's Scripts\\python.exe.
Restart the MCP host after changing its configuration.
Installation only installs packages. It does not initialize or migrate a workspace, deploy rules, modify Host instructions, or configure another application.
CLI — Setup, inspect, and diagnose
CLI = Setup + Observe + Diagnose; MCP = Work.
| Command | Purpose |
|---|---|
fcop init | Initialize a pure FCoP workspace |
fcop status | View workspace status |
fcop inspect | Inspect TASK / REPORT / ISSUE / REVIEW |
fcop validate | Validate protocol structure |
fcop tools | Inspect the installed MCP catalog |
fcop doctor | Diagnose installation, environment, and compatibility |
fcop version | Show installed versions |
fcop spec | Show specification and rule identity |
fcop migrate | Explicitly plan or apply a legacy workspace migration |
python -m pip install fcop
fcop version
fcop doctor
fcop init --root ./my-project
fcop status --root ./my-project
fcop validate --root ./my-project
python -m pip install fcop-mcp
fcop tools
fcop tools merge_branches --json
The local CLI operates without starting MCP. fcop tools reads this adapter's
canonical catalog without starting a server. Work operations such as
create_task, lifecycle transitions, Branch merge, and authorization use MCP
or the Python API. Canonical v4 workspace state belongs under <project>/fcop/.
fcop doctor does not access the network or modify Host configuration.
Authority and package boundary
An optional authorization Profile and evaluator must be supplied by the trusted application. A role name, TASK field, Profile document, or tool argument does not grant authority by itself.
| Package | Responsibility |
|---|---|
fcop | Protocol Core, validation, atomic persistence, and recovery |
fcop-mcp | 25 canonical MCP tools and 6 read-only Core resources |
FCoP uses local stdio by default. Relay support is an explicit optional extra.
Python applications that use Project directly only need fcop.
中文简介
fcop-mcp 4.0.5 是 FCoP Core 的薄 MCP 适配层。默认服务精确提供
25 个 Canonical MCP Tools和 6 个只读 Core Resources。
Core 是协议事实和状态迁移的唯一权威。Team、Solo、ME、seat、Host session、
Profile、Toolkit、Compatibility 与 Runtime 都不会成为 Base Protocol 的默认前置条件。
历史 49 工具实现面的每个入口均已在 4.0.5 迁移指南中给出明确去向。
Documentation
- Repository
- Authoritative install prompt (English)
- 权威安装提示词(中文)
- Installation prompt MCP Resource
- MCP tools reference
- 4.0.5 migration guide
- Protocol introduction
- 中文协议入门
- Changelog
License
MIT · LICENSE
Comments
New to MCP?
From what the protocol is to installing your first server, these answers cover most first questions.
An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.