Optiqra MCP server

optiqra-mcp

An MCP (Model Context Protocol) server that exposes every tool in OptiQra's API to MCP-compatible AI clients (Claude Desktop, Claude Code, Cursor, etc.).

OptiQra has four HTTP endpoints. This server wraps all four as MCP tools:

By default it talks to the public demo, https://optiqra.vercel.app. Point it at your own deployment (see OptiQra's own DEPLOYMENT.md/Docker setup) with an env var — see below.

Install

npm install

(or, once published, npm install -g optiqra-mcp / run directly with npx optiqra-mcp)

Configuration (environment variables)

Run it standalone

node src/index.js

It speaks MCP over stdio, so you won't see anything happen — it's waiting for an MCP client to connect.

Wire it into an MCP client

Claude Desktop / Claude Code

Add to your MCP config (claudedesktopconfig.json, or via claude mcp add for Claude Code):

{
  "mcpServers": {
    "optiqra": {
      "command": "node",
      "args": ["/absolute/path/to/optiqra-mcp/src/index.js"],
      "env": {
        "OPTIQRA_BASE_URL": "https://optiqra.vercel.app",
        "OPTIQRA_PROVIDER_API_KEY": "sk-..."
      }
    }
  }
}

Any other MCP-compatible client (Cursor, Windsurf, etc.) uses the same
`command`/`args`/`env` shape — check that client's docs for where the config
file lives.

## Notes on the AI-key tools

`optiqra_ai_fix`, `optiqra_ai_insights`, and `optiqra_ai_test` all need a
provider API key, exactly like pasting one into OptiQra's own UI — OptiQra's
server forwards it straight to the provider (OpenAI, Anthropic, Google, Groq,
OpenRouter, Mistral, DeepSeek, or xAI) and never stores it. This server does
the same: it never persists keys. Prefer setting `OPTIQRA_PROVIDER_API_KEY`
in your MCP client's env config over having the model pass a key through
chat.