MCP Perplexity Server

You can run the MCP Perplexity Server to ask questions to a model behind the Perplexity API. It supports both a local stdio-based setup and a remote SSE/http endpoint, giving you flexible options for development and deployment.

How to use

Choose your transport and use an MCP client to interact with the server. If you run locally, you can start the server in stdio mode and connect via the local process. If you prefer a remote-like setup, you can run the server in SSE mode and query it through its HTTP/SSE endpoint. Your client will send questions, and the server returns model responses from Perplexity.

How to install

Prerequisites you need before installing: Node.js, pnpm, and Docker.

Option 1: Quick setup via Docker Compose

services:
  mcp-sse-server:
    image: ghcr.io/sibbl/mcp-perplexity-api:main
    ports:
      - "8080:8080"
    restart: unless-stopped
    environment:
      - MCP_TRANSPORT=sse
      - MCP_PERPLEXITY_API_KEY=ENTER_YOUR_PERPLEXITY_API_KEY
      # - MCP_SSE_AUTH_HEADER_VALUE=Bearer some_token_which_is_required # enables optional bearer authentication against the MCP server with hard coded value

Save this configuration in a directory and run the following commands to bring the stack up and expose the SSE endpoint at http://localhost:8080/sse.

docker compose up

Option 2: Quick setup via Node.js (local development)

Install dependencies

pnpm install

Start the server in stdio mode (default) or SSE mode (alternative)

# with stdio transport
export MCP_TRANSPORT=stdio
pnpm start

# or with SSE transport
export MCP_TRANSPORT=sse
pnpm start

Additional sections

Configuration options you can set to customize the server behavior include API keys, default models, transport mode, and SSE settings. The important environment variables are listed below with their purposes and example values.