Setup MCP Server

This guide explains how to set up and use the BanyanDB MCP server from pre-built binaries or Docker images.

Prerequisites

  • Node.js 24.6.0+ installed (for binary usage)
  • BanyanDB running and accessible
  • MCP client (e.g., Claude Desktop, MCP Inspector, or other MCP-compatible clients)

Using Pre-built Binary

The MCP server binary is included in the BanyanDB release package. After extracting the release, you can find the MCP server in the mcp directory.

1. Verify Binary

cd mcp
node dist/index.js

The server starts in stdio mode by default and waits for MCP client connections. Verify BanyanDB is accessible before starting (see Verifying BanyanDB Connection).

2. Configure Environment Variables

Set the following environment variables:

  • BANYANDB_ADDRESS: BanyanDB server address (default: localhost:17900). The server auto-converts gRPC port (17900) to HTTP port (17913).
  • TRANSPORT: Transport mode. Default is stdio. Set http only when you want to expose the MCP server over HTTP.
  • MCP_HOST: HTTP listen address. Default is 127.0.0.1.
  • MCP_PORT: HTTP listen port. Default is 3000.
  • MCP_AUTH_TOKEN: Optional bearer token for HTTP mode. Required when MCP_HOST is not a loopback address.
  • MCP_MAX_BODY_BYTES: Maximum HTTP request body size. Default is 1048576 (1 MiB).
  • MCP_RATE_LIMIT_WINDOW_MS: Per-client HTTP rate limit window in milliseconds. Default is 60000.
  • MCP_RATE_LIMIT_MAX_REQUESTS: Maximum HTTP requests allowed per client in each rate-limit window. Default is 60.
  • MCP_RATE_LIMIT_MAX_CLIENTS: Maximum number of client IDs tracked by the in-memory HTTP rate limiter. Default is 10000.

Address formats:

  • localhost:17900 - Local BanyanDB
  • 192.168.1.100:17900 - Remote server
  • banyandb.example.com:17900 - Hostname

3. Configure MCP Client

Create a configuration file for your MCP client. For example, for MCP Inspector, see MCP client - Basic Setup

Using Docker Image

The MCP server is available as a Docker image for easy deployment.

1. Pull Docker Image

docker pull ghcr.io/apache/skywalking-banyandb-mcp:{COMMIT_ID}

2. Run the Container

docker run -d \
  --name banyandb-mcp \
  -e BANYANDB_ADDRESS=banyandb:17900 \
  ghcr.io/apache/skywalking-banyandb-mcp:{COMMIT_ID}

3. Configure MCP Client for Docker

When using Docker, configure your MCP client to connect to the container:

{
  "mcpServers": {
    "banyandb": {
      "command": "docker",
      "args": [
        "run",
        "--rm",
        "-i",
        "-e", "BANYANDB_ADDRESS=banyandb:17900",
        "--network", "host",
        "ghcr.io/apache/skywalking-banyandb-mcp:{COMMIT_ID}"
      ]
    }
  }
}

4. Docker Compose Example

You can also use Docker Compose to run both BanyanDB and the MCP server together:

services:
  banyandb:
    image: ghcr.io/apache/skywalking-banyandb:{COMMIT_ID} # apache/skywalking-banyandb:{COMMIT_ID}
    container_name: banyandb
    command: standalone
    ports:
      - "17912:17912"  # gRPC port
      - "17913:17913"  # HTTP port
    volumes:
      - ./banyandb-data:/data

  mcp:
    image: ghcr.io/apache/skywalking-banyandb-mcp:{COMMIT_ID} # apache/skywalking-banyandb-mcp:{COMMIT_ID}
    container_name: banyandb-mcp
    environment:
      - BANYANDB_ADDRESS=banyandb:17900
    depends_on:
      - banyandb
    networks:
      - default

Configuration Options

Environment Variables

Variable Required Default Description
BANYANDB_ADDRESS No localhost:17900 BanyanDB server address. Auto-converts gRPC port (17900) to HTTP port (17913).
TRANSPORT No stdio Transport mode: stdio for standard I/O (default), http for Streamable HTTP.
MCP_HOST No 127.0.0.1 HTTP listen address. Only used when TRANSPORT=http.
MCP_PORT No 3000 HTTP port to listen on. Only used when TRANSPORT=http.
MCP_AUTH_TOKEN Conditionally unset Optional bearer token for HTTP mode. Required when MCP_HOST is not a loopback address.
MCP_MAX_BODY_BYTES No 1048576 Maximum HTTP request body size in bytes.
MCP_RATE_LIMIT_WINDOW_MS No 60000 Per-client HTTP rate-limit window in milliseconds.
MCP_RATE_LIMIT_MAX_REQUESTS No 60 Maximum HTTP requests allowed per client during each rate-limit window.
MCP_RATE_LIMIT_MAX_CLIENTS No 10000 Maximum number of client IDs retained by the in-memory HTTP rate limiter before oldest entries are evicted.

Verifying BanyanDB Connection

Before using the MCP server, verify that BanyanDB is running and accessible:

# Check HTTP endpoint
curl http://localhost:17913/api/healthz
# Should return: {"status":"SERVING"}

# Check gRPC endpoint (if grpcurl is installed)
grpcurl -plaintext localhost:17912 list

HTTP Transport Mode

By default the MCP server communicates over standard I/O (TRANSPORT=stdio), which is suitable for desktop clients such as Claude Desktop. Set TRANSPORT=http to expose the server as an HTTP endpoint instead. This mode uses the MCP Streamable HTTP transport, which supports both streaming (SSE) and direct JSON responses.

Start in HTTP Mode

# Default loopback bind
TRANSPORT=http BANYANDB_ADDRESS=localhost:17900 node dist/index.js

# Custom host and port
TRANSPORT=http MCP_HOST=127.0.0.1 MCP_PORT=8080 BANYANDB_ADDRESS=localhost:17900 node dist/index.js

# Non-loopback host requires an auth token
TRANSPORT=http \
MCP_HOST=0.0.0.0 \
MCP_PORT=3000 \
MCP_AUTH_TOKEN="$(openssl rand -hex 32)" \
BANYANDB_ADDRESS=localhost:17900 \
node dist/index.js

The server prints the listening address on startup:

BanyanDB MCP HTTP server listening on 127.0.0.1:3000/mcp

The single endpoint is POST /mcp. Other HTTP methods on /mcp return 405 Method Not Allowed, and all other paths return 404.

If MCP_AUTH_TOKEN is configured, clients must send:

Authorization: Bearer <token>

HTTP mode also enforces request-size and per-client rate limits using MCP_MAX_BODY_BYTES, MCP_RATE_LIMIT_WINDOW_MS, MCP_RATE_LIMIT_MAX_REQUESTS, and MCP_RATE_LIMIT_MAX_CLIENTS.

Configure an MCP Client for HTTP

MCP clients that support the Streamable HTTP transport (e.g. MCP Inspector, custom integrations) connect via a URL:

{
  "mcpServers": {
    "banyandb": {
      "url": "http://localhost:3000/mcp",
      "headers": {
        "Authorization": "Bearer replace-with-your-token"
      }
    }
  }
}

If you keep the default loopback bind and do not set MCP_AUTH_TOKEN, the headers block is not required.

Docker with HTTP Mode

When running in a container, expose the HTTP port and set the transport env var:

docker run -d \
  --name banyandb-mcp \
  -p 3000:3000 \
  -e TRANSPORT=http \
  -e MCP_HOST=0.0.0.0 \
  -e MCP_PORT=3000 \
  -e MCP_AUTH_TOKEN=replace-with-a-strong-random-token \
  -e BANYANDB_ADDRESS=banyandb:17900 \
  ghcr.io/apache/skywalking-banyandb-mcp:{COMMIT_ID}

Docker Compose example with HTTP mode:

services:
  banyandb:
    image: ghcr.io/apache/skywalking-banyandb:{COMMIT_ID}
    container_name: banyandb
    command: standalone
    ports:
      - "17912:17912"
      - "17913:17913"

  mcp:
    image: ghcr.io/apache/skywalking-banyandb-mcp:{COMMIT_ID}
    container_name: banyandb-mcp
    environment:
      - TRANSPORT=http
      - MCP_HOST=0.0.0.0
      - MCP_PORT=3000
      - MCP_AUTH_TOKEN=replace-with-a-strong-random-token
      - BANYANDB_ADDRESS=banyandb:17900
    ports:
      - "3000:3000"
    depends_on:
      - banyandb

Choosing a Transport

stdio (default) http
Best for Claude Desktop, local CLI clients Web integrations, remote clients, custom apps
Connection Subprocess stdin/stdout HTTP POST /mcp
Session state Persistent (single process) Stateless (new context per request)

Troubleshooting

Connection refused:

  • Verify BanyanDB is running: curl http://localhost:17913/api/healthz
  • Check BANYANDB_ADDRESS environment variable
  • Verify ports 17900 (gRPC) and 17913 (HTTP) are accessible
  • For Docker, ensure containers are on the same network

“Command not found: node”:

  • Install Node.js 24.6.0+ from nodejs.org
  • Or use Docker image instead

HTTP mode exits at startup:

  • If MCP_HOST is not 127.0.0.1, localhost, or ::1, set MCP_AUTH_TOKEN
  • Verify MCP_PORT, MCP_MAX_BODY_BYTES, MCP_RATE_LIMIT_WINDOW_MS, MCP_RATE_LIMIT_MAX_REQUESTS, and MCP_RATE_LIMIT_MAX_CLIENTS are positive integers

MCP server not appearing in client:

  • Verify JSON config is valid
  • Use absolute path to binary
  • Check environment variables are set correctly
  • Restart your MCP client after configuration changes

Next Steps