Glasses / Getting started

Connect Glasses.

Give your agent a catalogue of useful work to start from.

Use the hosted MCP

No Glasses account, API key or local server needed. Add this URL to a client that supports Streamable HTTP:

https://lcfr.ai/api/mcp

Transport Streamable HTTPAuthentication None

Choose your client

Run this in your terminal, then reload your client’s MCP connection.

codex mcp add glasses_public --url https://lcfr.ai/api/mcp

The connection is named glasses_public. Ask your agent to search Glasses before building.

In LiteLLM, open MCP Servers → Import MCP Connectors (or Add MCP Server → Import from JSON).

{
  "mcpServers": {
    "glasses_public": {
      "type": "http",
      "url": "https://lcfr.ai/api/mcp"
    }
  }
}

Choose HTTP transport and No Auth. Test the connection, then enable the server for the key or team your agent uses.

Download JSON

Using config.yaml?

Merge this into your existing proxy configuration and restart LiteLLM.

mcp_servers:
  glasses_public:
    url: https://lcfr.ai/api/mcp
    transport: http
    auth_type: none

Merge this entry into your client’s existing MCP configuration. Formats vary; for a form-based client, use the URL above with Streamable HTTP and no authentication.

{
  "mcpServers": {
    "glasses_public": {
      "url": "https://lcfr.ai/api/mcp"
    }
  }
}

Download JSON

Tests the public endpoint. No model call.

Glasses provides read-only search, candidate details and source evidence. Your agent uses its own model allowance to work with the results.

Install locally

Use the local app for private research, source trials and the shared workbench. You’ll need Git and Node.js 24 or newer.

git clone https://github.com/lcfrai/glasses.git
cd glasses
node scripts/setup.mjs
npm start

Open localhost:4317 and keep the terminal running. Setup asks before importing the public catalogue.

Connect your local app to Codex

In a second terminal, inside the Glasses folder:

npm run mcp:install

This registers glasses_local for private research and workspace operations. For other clients, run npm run mcp:config.

Optional AI processing uses your selected provider. Codex uses your account allowance; Jev has separate API billing.

Shared catalogue

Fetch the latest public snapshot from your local Glasses folder:

npm run catalogue:fetch

The snapshot contains public source metadata, citations and attributed assessments. Your local notes and workspaces stay on your machine.

Download catalogue View coverage

Troubleshooting

The connection times out

Check that your client or LiteLLM host can reach lcfr.ai. A cold first catalogue lookup may take longer; check your client’s MCP timeout setting and try again.

The server connects, but my agent cannot use it

Reload the MCP connection. In LiteLLM, also enable the server for your agent’s key or team. Tool discovery should show glasses_status, glasses_search, glasses_inspect and glasses_evidence.

Which URL and authentication should I use?

Use https://lcfr.ai/api/mcp with Streamable HTTP (called HTTP in LiteLLM) and no authentication. The catalogue JSON download is a different endpoint.

Explore the catalogue