claude-mcp-bridge

MCP server that lets any agent or MCP host (Cursor, Codex, Antigravity, another Claude Code, your own app) delegate to Claude Code running headless (claude -p). Claude is the strongest model here, so the tool surface is deliberately narrow — reasoning, review and autonomous work. Cheap mechanical work (mapping a repo, scanning files) is intentionally left out: it would burn expensive tokens, and Claude already explores natively (Explore subagent + LSP/Grep/Glob) inside delegate. For cheap exploration use a lighter bridge (e.g. cursor-mcp-bridge).

Mirrors the ergonomics of agy-bridge / codex: every tool takes an optional model and effort, returns a session_id, and supports follow_up to continue without resending context.

Tools

Every tool accepts: cwd (project root), model (e.g. opus, sonnet, haiku), effort (low \| medium \| high).

Requirements

• Node ≥ 18
• claude CLI installed and authenticated (claude once to log in).

Install

git clone https://github.com/JaimeJunr/claude-mcp-bridge.git
cd claude-mcp-bridge
npm install
npm run build

Register in an MCP host

Claude Code:

claude mcp add claude-bridge -s user -- node /abs/path/to/claude-mcp-bridge/dist/index.js

Cursor / Codex / any host — add to its mcp.json:

{
  "mcpServers": {
    "claude-bridge": {
      "command": "node",
      "args": ["/abs/path/to/claude-mcp-bridge/dist/index.js"]
    }
  }
}

Configuration (env)

Security: acceptEdits lets delegated Claude edit files autonomously; bypassPermissions additionally allows unrestricted shell. Only raise this for trusted workspaces. The review / analyze / search tools instruct Claude to stay read-only regardless.

Develop

npm test       # vitest — unit tests for arg building / json parsing
npm run dev    # run from source via tsx

License

MIT