MCP Integration

What is MCP?

The Model Context Protocol (MCP) is an initiative to standardize how AI models interact with data sources and tools. By running local MCP servers, your AI client can connect via stdio and execute external commands, interact with databases, or read from third-party APIs.

Connecting Servers

Parallax includes a fully integrated MCP client. It discovers and connects to servers automatically on load.

To configure your servers, create a JSON file at ~/.parallax/mcp-config.json with the following structure:

{
  "mcpServers": {
    "postgres": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-postgres", "postgresql://localhost/mydb"]
    },
    "github": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"],
      "env": {
        "GITHUB_TOKEN": "ghp_..."
      }
    }
  }
}

Server Configuration Schema

Field	Type	Required	Description
`command`	`string`	✅	The executable to spawn (e.g. `npx`, `node`, `python`)
`args`	`string[]`	—	CLI arguments passed to the command
`env`	`object`	—	Environment variables required by the server

How Tools are Mapped

When Parallax boots up:

It reads your mcp-config.json and spawns each server as a child process.
It lists all available tools from the server using the official @modelcontextprotocol/sdk.
It maps each tool into a native Gemini function-calling JSON block.

To prevent collisions with our native internal tools (like listDirectory and readFile), Parallax prefixes all MCP tools with their server name:

{serverName}_{tool.name}

For example, if you connect a github server that exposes a search_repos tool, the agent will see it mapped as github_search_repos.

Usage

You do not need to invoke MCP tools manually. They are automatically injected into the ToolLoopAgent's knowledge base exactly like native tools.

If you say "Search for the latest PRs in the open-source repo", Parallax will intelligently trigger github_search_prs and weave the fetched data into its context window.