MCP Server Endpoint

QAnswer serves the remote MCP servers configured in it as standard MCP endpoints, over Streamable HTTP. Any MCP client — Claude Code, Cursor, VS Code, an MCP SDK — can use a server's tools with a QAnswer API key: QAnswer checks the key and passes each call on to the server.

It is the MCP counterpart of the Chat Completions endpoint, which does the same for LLM providers.

Endpoint

Each server has its own URL, ending with the server's identifier. The server's API tab in QAnswer shows it, ready to copy.

POST https://app.qanswer.ai/backend/api/v1/mcp/YOUR_SERVER_UID

Authentication

Send a QAnswer API key with every request, in either header:

Authorization: Bearer <key> — the header MCP clients send by default.
QAnswer-Api-Key: <key>

Create a key on the API Access page. Your plan has to include API access.

The key stays in QAnswer: it is never sent to the MCP server.

Which servers a key reaches

A user, team or organization key reaches every MCP server its owner can use in QAnswer.
A key bound to one MCP server reaches that server only. Such a key is created through the API, with POST /api/api-key/resource/mcp/{id}, by someone who can at least read the server.
A key bound to an AI assistant or to an LM provider is refused.
Only servers whose authentication is set to None are served: a server set to API Key or OAuth answers 400.
QAnswer's built-in servers — Web, Image Generation, Code Execution, Document Search, Assistants and Zimbra — run inside QAnswer and are never served here.

Connect a client

Replace the URL with the one from the server's API tab, and YOUR_API_KEY with your key.

In claude_desktop_config.json (Settings → Developer → Edit Config), through mcp-remote, which runs on your machine and forwards to the endpoint. It needs Node.js 18 or later; restart Claude Desktop after saving:

{
  "mcpServers": {
    "qanswer-my-server": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://app.qanswer.ai/backend/api/v1/mcp/YOUR_SERVER_UID",
        "--header",
        "Authorization:${AUTH_HEADER}"
      ],
      "env": { "AUTH_HEADER": "Bearer YOUR_API_KEY" }
    }
  }
}

In a terminal. Add --scope user to use the server in every project:

claude mcp add --transport http qanswer-my-server "https://app.qanswer.ai/backend/api/v1/mcp/YOUR_SERVER_UID" \
  --header "Authorization: Bearer YOUR_API_KEY"

In ~/.cursor/mcp.json, or .cursor/mcp.json in a project; restart Cursor after saving:

{
  "mcpServers": {
    "qanswer-my-server": {
      "url": "https://app.qanswer.ai/backend/api/v1/mcp/YOUR_SERVER_UID",
      "headers": { "Authorization": "Bearer YOUR_API_KEY" }
    }
  }
}

For GitHub Copilot in VS Code, in ~/.copilot/mcp-config.json, or .mcp.json at the root of the workspace:

{
  "mcpServers": {
    "qanswer-my-server": {
      "type": "http",
      "url": "https://app.qanswer.ai/backend/api/v1/mcp/YOUR_SERVER_UID",
      "headers": { "Authorization": "Bearer YOUR_API_KEY" },
      "tools": ["*"]
    }
  }
}

In a terminal. -s user adds the server to ~/.gemini/settings.json, for every project:

gemini mcp add -s user --transport http \
  --header "Authorization: Bearer YOUR_API_KEY" qanswer-my-server "https://app.qanswer.ai/backend/api/v1/mcp/YOUR_SERVER_UID"

With version 2 of the official mcp package:

# pip install "mcp>=2"
import asyncio

import httpx2
from mcp import Client
from mcp.client.streamable_http import streamable_http_client

URL = "https://app.qanswer.ai/backend/api/v1/mcp/YOUR_SERVER_UID"
HEADERS = {"Authorization": "Bearer YOUR_API_KEY"}


async def main() -> None:
    async with httpx2.AsyncClient(
        headers=HEADERS, timeout=httpx2.Timeout(30.0, read=300.0)
    ) as http:
        async with Client(streamable_http_client(URL, http_client=http)) as client:
            result = await client.list_tools()
            for tool in result.tools:
                print(tool.name, "-", tool.description)
            # result = await client.call_tool("TOOL_NAME", {"argument": "value"})


asyncio.run(main())

With version 2 of the official MCP TypeScript SDK, on Node.js:

// npm install @modelcontextprotocol/client
// an ES module: save as client.mjs, then run node client.mjs
import {
  Client,
  StreamableHTTPClientTransport,
} from '@modelcontextprotocol/client'

const transport = new StreamableHTTPClientTransport(new URL('https://app.qanswer.ai/backend/api/v1/mcp/YOUR_SERVER_UID'), {
  requestInit: { headers: { Authorization: 'Bearer YOUR_API_KEY' } },
})
const client = new Client({ name: 'my-app', version: '1.0.0' })

await client.connect(transport)
const { tools } = await client.listTools()
for (const tool of tools) console.log(tool.name, '-', tool.description)
// const result = await client.callTool({ name: 'TOOL_NAME', arguments: { argument: 'value' } })
await client.close()

curl -X POST "https://app.qanswer.ai/backend/api/v1/mcp/YOUR_SERVER_UID" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -d '{"jsonrpc": "2.0", "id": 1, "method": "tools/list"}'

curl -X POST "https://app.qanswer.ai/backend/api/v1/mcp/YOUR_SERVER_UID" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -d '{"jsonrpc": "2.0", "id": 2, "method": "tools/call", "params": {"name": "TOOL_NAME", "arguments": {}}}'

Limitations

Tools only: resources, prompts and completions are not offered.
The tools offered are the ones synced into QAnswer and switched on by the server's owner, with the descriptions set in QAnswer.
The endpoint is stateless and answers in JSON: no session, no event stream, no progress notifications. A tool call answers once the MCP server has answered.
QAnswer's ask-for-confirmation setting applies to its own chat only; approving tool calls is up to your client.
Tool calls are not charged credits.
Clients that connect to remote servers only through OAuth, such as ChatGPT, cannot use the endpoint, which takes a static API key.

Errors

Problems with the request are HTTP errors. Problems with a single call come back in a normal 200 response, so the client keeps its connection.

401 / 403 — no key, an invalid or expired key, a key bound to another resource, or no access to the server
404 — no server with that identifier
400 — a built-in server, a server that requires authentication, a body that is not a single JSON-RPC message, or an unsupported MCP-Protocol-Version header
200, JSON-RPC -32601 — a method other than initialize, ping, tools/list and tools/call
200, JSON-RPC -32602 — invalid parameters, or a tool the server does not offer
200, isError — the tool failed, or the MCP server could not be reached

MCP Server Endpoint#

Endpoint#

Authentication#

Which servers a key reaches#

Connect a client#

Limitations#

Errors#