MCP Observatory

Instructions

Implementation Reference

• src/server.ts:72-99 (handler)

  The `check_server` MCP tool is defined here, which validates input arguments and calls `runTarget` to perform the checks.

  server.tool(
    "check_server",
    "Run checks against a specific MCP server by command. Example: check_server({ command: 'npx -y @modelcontextprotocol/server-everything' })",
    {
      command: z.string().describe("The command to launch the MCP server (e.g. 'npx -y @modelcontextprotocol/server-everything')."),
      args: z.array(z.string()).optional().describe("Additional arguments for the command."),
    },
    async ({ command, args }) => {
      try {
        const target = {
          targetId: command,
          adapter: "local-process" as const,
          command,
          args: args ?? [],
          timeoutMs: 15_000,
        };
        const artifact = await runTarget(target);
        const outDir = defaultRunsDirectory();
        const outPath = await writeRunArtifact(artifact, outDir);
        return {
          content: [{ type: "text" as const, text: `${formatRun(artifact)}\n\nArtifact saved: ${outPath}` }],
        };
      } catch (error) {
        const msg = error instanceof Error ? error.message : String(error);
        return { content: [{ type: "text" as const, text: `Error checking server: ${msg}` }], isError: true };
      }
    },
  );