Clawy MCP Server

Instructions

Implementation Reference

• src/index.ts:20-39 (handler)

  The generic tool handler in src/index.ts executes the llm_chat tool by making an HTTP request to the endpoint defined in the tool configuration (src/llm/chat.ts).

    async (params) => {
      const method = tool.method || "POST";
      const result = await gatewayRequest(method, tool.endpoint, params as Record<string, unknown>);
  
      if (result.error) {
        return {
          content: [{ type: "text" as const, text: `Error (${result.status}): ${result.error}` }],
          isError: true,
        };
      }
  
      const text = typeof result.data === "string"
        ? result.data
        : JSON.stringify(result.data, null, 2);
  
      return {
        content: [{ type: "text" as const, text }],
      };
    },
  );

• src/llm/chat.ts:8-19 (schema)

  Schema definition for llm_chat input parameters using Zod.

    inputSchema: z.object({
      model: z.enum(["auto", "gpt-5-nano", "kimi-k2p5", "claude-opus-4-6"]).optional()
        .describe("Model selection: 'auto' for smart routing (recommended), or specify a model directly"),
      messages: z.array(z.object({
        role: z.enum(["system", "user", "assistant"]),
        content: z.string(),
      })).describe("Chat messages array"),
      temperature: z.number().optional().describe("Sampling temperature (0-2, default 0.7)"),
      max_tokens: z.number().optional().describe("Maximum response tokens"),
    }),
    endpoint: "/v1/llm/chat",
  },

• src/llm/chat.ts:6-19 (registration)

  Registration and configuration of the llm_chat tool.

    name: "llm_chat",
    description: "Smart-routed LLM chat completion. Automatically selects the optimal model (Claude, GPT, Gemini, Llama) based on task complexity. No API keys needed — pay per call with USDC credits.",
    inputSchema: z.object({
      model: z.enum(["auto", "gpt-5-nano", "kimi-k2p5", "claude-opus-4-6"]).optional()
        .describe("Model selection: 'auto' for smart routing (recommended), or specify a model directly"),
      messages: z.array(z.object({
        role: z.enum(["system", "user", "assistant"]),
        content: z.string(),
      })).describe("Chat messages array"),
      temperature: z.number().optional().describe("Sampling temperature (0-2, default 0.7)"),
      max_tokens: z.number().optional().describe("Maximum response tokens"),
    }),
    endpoint: "/v1/llm/chat",
  },