Skip to content

windsurf.md

Latest commit

 

History

History
53 lines (35 loc) · 1.63 KB

windsurf.md

File metadata and controls

53 lines (35 loc) · 1.63 KB

CodeTalker for Windsurf

Windsurf (formerly Codeium) supports MCP via Cascade — its agent integrates with MCP servers from Wave 3 (Feb 2025) onward.

Install

1. Install the daemon

pip install --user claude-code-talker

2. Add to Windsurf's MCP config

Edit ~/.codeium/windsurf/mcp_config.json:

{
  "mcpServers": {
    "codetalker": {
      "command": "codetalker-mcp",
      "args": [],
      "env": {}
    }
  }
}

Or use Windsurf's Settings → MCP Servers → "Add Server" UI:

  • Name: codetalker
  • Command: codetalker-mcp
  • Args: (empty)

3. Restart Windsurf

The MCP server is loaded at Cascade startup. Restart Windsurf to pick up the new config.

4. Cascade and the 100-tool ceiling

Windsurf enforces a hard limit of 100 tools per Cascade session. CodeTalker contributes 9 agent-facing tools (plus 5 internal hook handlers not exposed via the shim). If you have many other MCP servers attached, prioritize codetalker on the "active" list in MCP settings.

Verify

In a Cascade chat:

Call the codetalker tts_status tool to report state.

Look for a blue-checkmark badge on the codetalker server in MCP settings — that means Windsurf verified the server passes their basic compatibility check.

Notes specific to Windsurf

  • Cascade's "Write" mode involves longer agentic loops than its "Chat" mode. Switch codetalker to live mode (tts_set_mode live) when starting a Write task, then back to direct for chat.
  • If you toggle Cascade between local models and Anthropic Claude / OpenAI GPT, codetalker keeps working — it doesn't care which model is on the other end of MCP.