LABSCO SUMMARY

The MCP surface is content, not execution, and it says so.

An empty tools/list and a not_implemented envelope on tools/call is an unusual thing to document rather than paper over — execution is deferred and the local surface is read-only by design, so you know exactly what you are connecting. There is no public endpoint either: the Worker source ships, but you deploy it yourself. Start with the single read-only subagent before installing the suite; it needs nothing else.

What it is

A configuration layer for AI coding agents: a deep library of skills, commands, rules and personas, plus a capability router that loads the right skill on intent instead of dumping the whole catalog into context. It compiles into 20 host agents — Claude Code, Cursor, Augment, Cline, Windsurf, Copilot, Gemini CLI, Codex, Continue, Zed, JetBrains, Aider and more — and the MCP server is the read-only way to reach that content from any client.

What you get

A read-only content surface over MCP — skills and commands served as prompts, rules and guidelines as resources. The local stdio server lists no tools (tools/call returns a not_implemented envelope by design); the self-hosted Worker adds small read-only tools such as memory_lookup and chat_history_read
Two entry points: a local stdio server (agent-config mcp-server) that needs no account and no network, and a self-hosted Cloudflare Worker for a shared team endpoint
A 30-second starting point — a single read-only production-validator subagent dropped into any repository, which gates "done" by hunting mocks and stubs on the shipped path and demanding real-system evidence
Governance compiled into each host's native rule format at projection time, with deterministic runtime hooks added on hook-capable hosts
A detection-driven installer that finds your installed AI tools and pre-selects them, writing nothing until you click Finish
Six role-shaped entry paths — developer, founder, content, agency, finance, ops — so you get a focused command set rather than the whole library

Requirements

No account and no network for the local path: npm install -g @event4u/agent-config provides the agent-config binary, and the client entry is that command with the argument mcp-server — or npx -y @event4u/agent-config mcp-server if you would rather not install globally. Setup runs through npx -y @event4u/agent-config init, which launches a browser wizard on a terminal with a display and falls back to the non-interactive installer on CI, a non-TTY, a headless host, or when any CLI-mode flag is present; --profile and --tools pick your profile and hosts, and --dry-run previews writes. The remote path needs your own deployed Cloudflare Worker — the project ships the source but operates no shared hosted endpoint. Stdio-only clients reaching the Worker need the mcp-remote bridge.

Setup effort

One command — npx -y @event4u/agent-config mcp-server

See installation options

USE CASES

01

Trying it without committing to anything

Drop the read-only validator subagent into a repo and let it gate whether a branch is actually done.

02

Getting the same rules onto every tool your team uses

Run the installer with the tools flag listing the hosts you have, so the rules compile into each one's native format.

03

Reading the library from a client that only speaks MCP

Point it at the local stdio command and browse skills and commands as prompts, rules and guidelines as resources.

04

Installing on CI without a wizard

Pass the profile and tools flags — the GUI is skipped automatically on CI and on a non-TTY — and dry-run first to see what would be written.

2 tools

Listed in project docs. Your client may show a different list.

memory_lookup

chat_history_read

BEFORE YOU INSTALL

What to know first

Tools you get

2 tools named in the project's documentation. Both tools come from the self-hosted Cloudflare Worker; the local agent-config mcp-server lists no tools.

Requires

npx on PATH
Node ≥ 20.11

INSTALLATION

Connect your client

Server setup

TERMINAL

npx -y @event4u/agent-config mcp-server

MCP Clients

Claude Desktop, Cursor

TERMINAL

npm install -g @event4u/agent-config

This provides the agent-config binary the entry below runs (or, to skip the global install, use "command": "npx", "args": ["-y", "@event4u/agent-config", "mcp-server"] instead of "command": "agent-config", "args": ["mcp-server"]).

JSON

{
  "mcpServers": {
    "agent-config": {
      "command": "agent-config",
      "args": [
        "mcp-server"
      ]
    }
  }
}

Claude Code

CLAUDE CODE · TERMINAL

npm install -g @event4u/agent-config

This provides the agent-config binary. Then register it:

CLAUDE CODE · TERMINAL

claude mcp add agent-config -- agent-config mcp-server

Zed

ZED · TERMINAL

npm install -g @event4u/agent-config

This provides the agent-config binary the entry below runs (or, to skip the global install, use "command": "npx", "args": ["-y", "@event4u/agent-config", "mcp-server"] instead of "command": "agent-config", "args": ["mcp-server"]).

ZED · JSON

{
  "context_servers": {
    "agent-config": {
      "source": "custom",
      "command": "agent-config",
      "args": [
        "mcp-server"
      ]
    }
  }
}

Continue

The project documents Continue only against its self-hosted Cloudflare Worker, not the local agent-config server — and it runs no public endpoint. Deploy your own Worker first, then add this to ~/.continue/config.yaml (or <repo>/.continue/config.yaml):

CONTINUE · YAML

mcpServers:
  - name: agent-config
    command: npx
    args:
      - "-y"
      - mcp-remote
      - https://<YOUR_WORKER>.workers.dev
      - --header
      - "Authorization: Bearer ${MCP_TOKEN}"
    env:
      MCP_TOKEN: <YOUR_MCP_TOKEN>

https://<YOUR_WORKER>.workers.dev is your Worker's URL, and <YOUR_MCP_TOKEN> is the MCP token you set on it. Drop the --header and env lines when the Worker has no token set.

@event4u/agent-config