Connecting your editor¶

Your editor talks to Veska over MCP by launching veska-mcp - the thin stdio shim that proxies JSON-RPC frames to the running daemon's Unix socket. Point your MCP client at the absolute path to bin/veska-mcp.

The daemon must be running

veska-mcp is only a proxy. Start the daemon first - see Running the daemon.

Claude Desktop¶

~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "veska": {
      "command": "/abs/path/to/bin/veska-mcp"
    }
  }
}

Cursor, Zed, Continue¶

Cursor (~/.cursor/mcp.json) and Zed (~/.config/zed/settings.json, under context_servers) accept the same command shape as above.

Continue (~/.continue/config.yaml):

mcpServers:
  - name: veska
    command: /abs/path/to/bin/veska-mcp

Non-default `VESKA_HOME`¶

If your data root isn't ~/.veska, pass it through so the shim finds the right socket:

{
  "mcpServers": {
    "veska": {
      "command": "/abs/path/to/bin/veska-mcp",
      "env": { "VESKA_HOME": "/path/to/veska/home" }
    }
  }
}

Tell the agent the tools exist¶

veska init --agent <name> writes (or safely appends to) an agent-specific instruction file in the current project, so the agent knows the Veska tool surface is available. Supported: claude, codex, copilot, cursor, gemini, kiro, opencode.

cd /path/to/your/repo
veska init --agent claude    # creates or updates CLAUDE.md

The snippet is bracketed with  markers, so re-running updates only the Veska section and leaves the rest of the file alone.

Verify without an editor¶

You can drive veska-mcp straight from the shell - handy for debugging:

printf '{"jsonrpc":"2.0","id":1,"method":"eng_find_symbol","params":{"symbol":"Run"}}\n' \
  | ./bin/veska-mcp | jq '.result.nodes[0]'

See the MCP tools reference for the full tool surface.