@elyracode/mcp

Model Context Protocol client for Elyra. It uses the MCP servers you already configured for Claude Code, Cursor, VS Code, or Elyra, so there is nothing to copy.

Bundled with @elyracode/coding-agent: every Elyra install has it. With no servers configured it adds no tools and costs nothing. Turn it off with "mcp": false in ~/.elyra/agent/settings.json.

What it does

• Reads existing configuration from ~/.claude.json, ~/.cursor/mcp.json, ~/.elyra/agent/mcp.json, and, in trusted projects only, .mcp.json, .elyra/mcp.json, .cursor/mcp.json, and .vscode/mcp.json.
• Adds two stable tools instead of every server's schemas: mcp_search finds a tool, mcp_call runs it. The tools block never changes with the number of servers, which keeps the prompt cache warm and the context small.
• Starts servers lazily and caches tool lists on disk, so searching rarely needs a running server.
• Direct tools on request. "directTools": true on a server also registers each of its tools as mcp__<server>__<tool>. Tools that appear later are registered with defer, so they do not break the prompt cache mid-conversation.
• OAuth for remote servers via /mcp login <server>, with dynamic client registration, PKCE, and automatic refresh. Tokens are stored with 0600 permissions.
• Shares connections with child agents. Swarm stages reuse the parent session's servers through a local broker instead of starting their own.
• Keeps secrets in. Stdio servers get a minimal environment plus the env you configure, not your whole shell.

Commands

/mcp                       status of every server, plus skipped files and config problems
/mcp login <server>        OAuth login for a remote server
/mcp logout <server>       forget stored tokens
/mcp refresh [server]      reconnect and refetch tool lists
/mcp disable <server>      switch a server off
/mcp enable <server>
/mcp reload                re-read all configuration files

See docs/mcp.md for the configuration format and details.