home / mcp / perplexica mcp server

Perplexica MCP Server

Provides an MCP interface to local Perplexica searches with focus modes, streaming, and history.

Installation

Add the following to your MCP client configuration file.

Configuration

{
  "mcpServers": {
    "armand0e-perplexica-mcp": {
      "url": "http://localhost:3000",
      "headers": {
        "PERPLEXICA_BASE_URL": "http://localhost:3000"
      }
    }
  }
}

You can run a local MCP server that interfaces with your Perplexica installation to perform AI-powered searches, streaming results, and managed conversations. This server exposes AI-powered search capabilities to your MCP clients and keeps model and conversation context handy for richer interactions.

How to use

You connect to the Perplexica MCP server from your MCP client by configuring a server entry that points to your local Perplexica instance. Run the server either as a local process (stdio) or as a remote HTTP endpoint. With the stdio approach, you launch the Python module that runs the MCP logic; with the HTTP approach, you point your client at a local URL where the server is listening.

Key capabilities you can leverage include AI-powered searches across different focus modes, real-time streaming responses, and the ability to preserve conversation history across searches. Tune the base URL so your client talks to the correct Perplexica instance, and ensure the environment variable that defines the base URL matches where Perplexica is running.

How to install

Prerequisites you need on your machine: Python 3.8 or newer, Git, and an internet connection to install dependencies.

git clone https://github.com/armand0e/perplexica-mcp.git
cd perplexica-mcp
pip install -e .

Configuration and usage notes

The server connects to a local Perplexica instance. By default, Perplexica is expected to run at http://localhost:3000. You can override this with the PERPLEXICA_BASE_URL environment variable.

Example to set the base URL for Perplexica to a different port if needed:

export PERPLEXICA_BASE_URL=http://localhost:3000

Run via stdio (local process)

Launch the MCP server as a Python module in a local process. This makes the server available to MCP clients via a local standard input/output stream.

python -m perplexica_mcp

Run via http (remote/local HTTP endpoint)

If you prefer exposing the server as an HTTP endpoint, point your MCP client to the local Perplexica URL. Ensure Perplexica is running at the configured address.

Available tools

perplexica_search

Performs AI-powered searches with various focus modes and returns results, with options for streaming responses.

perplexica_get_models

Retrieves available chat and embedding models supported by the Perplexica MCP integration.