I've connected vapi to our own MCP (based on @vercel/mcp-adapter package), which works perfectly fine for many other use cases so far. It seems to connect, but I see no activity after the initial connection. And the agent seems unable to do anything in relation to the MCP server. On start of conversation I see these logs on my server:

{
  type: 'REQUEST_COMPLETED',
  method: 'initialize',
  result: {
    method: 'initialize',
    params: {
      protocolVersion: '2025-03-26',
      capabilities: [Object],
      clientInfo: [Object]
    },
    jsonrpc: '2.0',
    id: 0
  },
  duration: 4,
  status: 'success',
  sessionId: undefined,
  requestId: 'c3dbee07-d10e-4a52-adfa-290f00334b31',
  timestamp: 1754560605791
}
{
  type: 'REQUEST_COMPLETED',
  method: 'tools/list',
  result: { method: 'tools/list', jsonrpc: '2.0', id: 1 },
  duration: 2,
  status: 'success',
  sessionId: undefined,
  requestId: 'e5002aa0-2c7a-4acd-ac63-b53c320064ac',
  timestamp: 1754560606618
}

These then repeat at the END of the conversation. But there is then ZERO activity on the server at all when I try to trigger the calls (even explicitely). I've tried both SSE and SHTTP but the same behaviour happens for both. A reference call ID: 43849ee5-6eea-47c3-8589-429a9045f299

Suggestion: Check that your MCP server's

tools/list

response includes a valid array of tool definitions with proper schemas, as Vapi requires these to inject tools into the assistant's context. For example, your response should look like:

json
{
  "jsonrpc": "2.0",
  "id": 1,
  "result": {
    "tools": [
      {
        "name": "your_function",
        "description": "What this function does",
        "parameters": { "type": "object", "properties": { /* ... */ } }
      }
    ]
  }
}

If the tools array is empty or missing required fields, the assistant will not be able to call any functions. See [Vapi MCP Tool documentation](https://docs.vapi.ai/tools/mcp#how-mcp-works) for required structure. Source: - [Model Context Protocol (MCP) Integration](https://docs.vapi.ai/tools/mcp#how-mcp-works)

I've escalated your issue to our executives for their attention. Thank you for your patience, we appreciate your understanding and will keep you updated..