Tool Execution & Errors — Tools & Functions

Handling Tool Errors Gracefully

When an LLM provides bad arguments or an API call fails, your tool shouldn't crash the server. It should return a graceful error message back to the LLM so the AI can debug itself and try again.

server.registerTool(
  "read_file",
  {
    description: "Reads a UTF-8 file from disk",
    inputSchema: z.object({ path: z.string().describe("Target file path") })
  },
  async ({ path }) => {
    try {
      const data = await fs.readFile(path, 'utf8');
      return { content: [{ type: "text", text: data }] };
    } catch (e: any) {
      // ✅ Allow the LLM to learn and retry:
      return { 
        isError: true, 
        content: [{ type: "text", text: `Error reading file. Did you use the correct path? ${e.message}`}] 
      };
    }
  }
);