tl-openmeter-api-mcp-server（TL openmeter API MCP server 搜索）

tl-openmeter-api-mcp-server

MCP server for your local OpenMeter instance: tools and resources so AI assistants (Cursor, Claude, etc.) can list meters, manage customers and subscriptions, query usage, and read the API quick reference.

Pair this with the tl-openmeter-api skill for knowledge; use this server to run operations against your OpenMeter.

Suite

When to Use

• "List my OpenMeter meters"
• "Create a customer in OpenMeter"
• "Query usage for this meter"
• "Ingest a test event"
• "Check if OpenMeter is running"
• When you need the agent to actually call OpenMeter API (not just provide documentation)

Prerequisites

• Node.js 18+
• A running OpenMeter instance (e.g. Docker, default http://localhost:8888)

Installation

Step 1: Build the Server

From the skill directory:

npm install
npm run build

Step 2: Register in Cursor MCP Config

Add to ~/.cursor/mcp.json:

{
  "mcpServers": {
    "OpenMeter": {
      "command": "node",
      "args": ["C:\\Users\\Todd\\.cursor\\skills\\@tl-agent-skills\\skills\\tl-openmeter-api-mcp-server\\dist\\index.js"],
      "env": {
        "OPENMETER_URL": "http://localhost:8888",
        "OPENMETER_API_KEY": ""
      }
    }
  }
}

Adjust the path to match your installation location.

Step 3: Restart Cursor

Cursor needs to restart to pick up MCP config changes.

Environment Variables

Local dev often runs without OPENMETER_API_KEY.

Available Tools

Resources

• OpenMeter quick reference — Markdown from the tl-openmeter-api skill
• OpenMeter spec note — Plain text with spec URL and configured base URL

Verification

1. Start OpenMeter (e.g. docker compose up or your stack)
2. Ensure the OpenMeter MCP server is configured and enabled in Cursor
3. In a chat, ask: "Check the OpenMeter status" or "List OpenMeter meters"
4. The agent should call openmeter_check_status or openmeter_list_meters and return results

Errors (e.g. connection refused) are returned as tool results so the agent can report them.

Troubleshooting

MCP Design Principles

From anthropics/skills/mcp-builder:

Tool Naming

Use consistent, action-oriented prefixes:

openmeter_list_meters     ✓ (list action)
openmeter_get_customer    ✓ (get action)
openmeter_create_subscription ✓ (create action)
meters                    ✗ (no action prefix)

API Coverage vs Workflow Tools

Balance comprehensive API endpoint coverage with specialized workflow tools:

• Comprehensive coverage gives agents flexibility to compose operations
• Workflow tools (e.g., openmeter_provision_customer) combine multiple calls for common tasks

When uncertain, prioritize comprehensive API coverage.

Actionable Error Messages

Error messages should guide agents toward solutions:

// Bad
throw new Error('Request failed');

// Good
throw new Error(
  `OpenMeter customer not found: ${customerId}. ` +
  `Use openmeter_list_customers to find valid customer IDs.`
);

Pagination

All list operations support cursor-based pagination:

interface PaginationParams {
  page?: number;
  pageSize?: number;  // default: 100, max: 1000
}

interface PaginatedResponse<T> {
  items: T[];
  page: number;
  pageSize: number;
  totalCount: number;
}

Tool Implementation

server.registerTool({
  name: 'openmeter_list_customers',
  description: 'List customers with pagination. Returns page info and total count.',
  inputSchema: z.object({
    page: z.number().optional().describe('Page number (1-indexed)'),
    pageSize: z.number().max(1000).optional().describe('Items per page, max 1000'),
  }),
  async handler({ page = 1, pageSize = 100 }) {
    const response = await fetch(
      `${baseUrl}/api/v1/customers?page=${page}&pageSize=${pageSize}`
    );
    return {
      content: [{ type: 'text', text: JSON.stringify(await response.json()) }],
    };
  },
});

Batch Operations

Batch Event Ingestion

server.registerTool({
  name: 'openmeter_batch_ingest',
  description: 'Ingest multiple CloudEvents in a single request',
  inputSchema: z.object({
    events: z.array(z.object({
      type: z.string(),
      subject: z.string(),
      data: z.record(z.unknown()),
    })),
  }),
  async handler({ events }) {
    const cloudEvents = events.map(e => ({
      specversion: '1.0',
      id: crypto.randomUUID(),
      source: 'mcp-server',
      type: e.type,
      subject: e.subject,
      time: new Date().toISOString(),
      data: e.data,
    }));

    const response = await fetch(`${baseUrl}/api/v1/events`, {
      method: 'POST',
      headers: { 'Content-Type': 'application/cloudevents-batch+json' },
      body: JSON.stringify(cloudEvents),
    });

    return { content: [{ type: 'text', text: `Ingested ${events.length} events` }] };
  },
});

Tool Annotations

Mark tools with hints for agent behavior:

server.registerTool({
  name: 'openmeter_create_customer',
  annotations: {
    readOnlyHint: false,      // Modifies state
    destructiveHint: false,   // Does not delete data
    idempotentHint: false,    // Creates new resource each time
    openWorldHint: true,      // External service interaction
  },
  // ...
});

server.registerTool({
  name: 'openmeter_list_meters',
  annotations: {
    readOnlyHint: true,       // Only reads data
    destructiveHint: false,
    idempotentHint: true,     // Same result for same input
    openWorldHint: true,
  },
  // ...
});

Testing

MCP Inspector

Test tools interactively:

npx @modelcontextprotocol/inspector

Mock Server for Tests

import { createMockMcpServer } from './test-utils';

const mockServer = createMockMcpServer({
  tools: {
    openmeter_list_meters: async () => ({
      content: [{ type: 'text', text: JSON.stringify([
        { id: 'test-meter', slug: 'api-calls', aggregation: 'COUNT' }
      ]) }],
    }),
  },
});

Integration Test Pattern

describe('OpenMeter MCP Server', () => {
  it('lists meters from running OpenMeter', async () => {
    const result = await server.callTool('openmeter_list_meters', {});
    expect(result.content[0].type).toBe('text');
    const meters = JSON.parse(result.content[0].text);
    expect(Array.isArray(meters)).toBe(true);
  });

  it('returns actionable error for invalid customer', async () => {
    const result = await server.callTool('openmeter_get_customer', {
      idOrKey: 'nonexistent'
    });
    expect(result.content[0].text).toContain('openmeter_list_customers');
  });
});

References

Quilted Skills

• anthropics/skills/mcp-builder — MCP server development guide

First-Party Documentation

• MCP Specification — Protocol specification
• MCP TypeScript SDK — SDK reference
• OpenMeter API — API endpoints

MCP Resources

• MCP Server Examples — Reference implementations
• MCP Inspector — Debugging tool
• MCP Best Practices — Universal guidelines