Cursor

Use ImplCache as a Cursor MCP server

ImplCache works with Cursor and other MCP clients over stdio or HTTP. The key is pointing Cursor at the binary with absolute paths and the right mode for the job.

Install a package

Grab a ready-to-run build from the GitHub repo:

Each package includes the server binary, ingestcli, an empty starter database, and helper scripts. No Go install is required to try it.

Point Cursor at the binary

In Cursor MCP settings (or mcp.json), use absolute paths:

{
  "mcpServers": {
    "implcache": {
      "command": "D:/path/to/dist/implcache-mcp.exe",
      "args": [
        "-db", "D:/path/to/dist/implcache.db",
        "-mode", "agent"
      ]
    }
  }
}

On Linux or Jetson, point command at the implcache-mcp binary in the matching package directory.

Agent mode vs admin mode

  • agent — retrieval for coding sessions.
  • admin — ingest, delete, and corpus maintenance tools.

Keep day-to-day Cursor coding on agent. Use admin mode (or the Librarian UI / ingestcli) when you are loading or updating sources.

How agents should call tools

  1. Start with get_implementation_context and a clear task, language, and preferred roots when known.
  2. If the package is incomplete, use find_symbol and search_knowledge.
  3. Call get_document only when you truly need the full source document.

That staged approach keeps prompts small and reduces “read the whole tree” behavior.

Optional Librarian UI

Launch the embedded Librarian to manage sources and inspect health:

# Windows
.\run-librarian.cmd

# Linux / Jetson
./run-librarian.sh
# open http://127.0.0.1:8080/

For a remote Jetson host, see Jetson Orin MCP server. For package shape details, see implementation context.

Ready to try it? Packages ship for Windows, Linux, and Jetson.

Run locally GitHub