Updated 22 June 2026

trovex for Windsurf

Windsurf keeps MCP servers in ~/.codeium/windsurf/mcp_config.json and manages them from Cascade. Add trovex and the agent gets one canonical doc per query — about 60% fewer tokens.

{
  "mcpServers": {
    "trovex": { "serverUrl": "http://localhost:8765/mcp" }
  }
}

trovex is open source and in public beta. The five steps below run in about a minute — no account, no keys.

Quickstart: from install to the tokens saved

1. Install trovex.
   uv tool install trovexcopy
2. Index your repo.

   Embeds your markdown locally (SQLite + ONNX). About a minute on a typical repo.

   trovex index /path/to/repocopy
3. Ask a question.
   trovex search "how do we roll back a deploy?"copy
4. See the tokens saved.

   trovex returns the single current doc that answers — down to the section — and prints how many tokens that saved versus rereading the top few candidates. That number is the point. Look for the green ≈ N tokens saved this query line that confirms it worked.

5. Add it to Windsurf.

   Start the server, add a trovex entry with serverUrl to ~/.codeium/windsurf/mcp_config.json (Windows: %USERPROFILE%\.codeium\windsurf\mcp_config.json), then refresh in Cascade → MCP / Plugins.

   trovex servecopy

   {
     "mcpServers": {
       "trovex": { "serverUrl": "http://localhost:8765/mcp" }
     }
   }

Older Windsurf builds connect to HTTP servers through the mcp-remote bridge instead of serverUrl. If trovex doesn't appear, check Windsurf's current MCP docs for the format your version expects.

Why Windsurf users want it

Cascade is strong at navigating code, but for project knowledge spread across many markdown files it still rereads to find the current one. trovex serves the single canonical doc that answers, with a freshness marker, so Cascade reads the right section instead of a stale copy — about 60% fewer tokens per lookup, with a shared write path so context carries across sessions.