Live · https://canonn.ai/mcp

Canonn for Claude Code, Codex and OpenCode.

Keep your primary coding model. Canonn becomes the evidence layer it calls when a task depends on internal documents or requires verifiable citations.

Available MCP tools

canonn_answer — answer from authorized sources and return passage-level citations.
canonn_list_sources — list the knowledge files the supplied key can read.
canonn_get_source — retrieve a knowledge file's metadata by file id.

Connect Claude Code

Create a Canonn API key in the dashboard, then run:

claude mcp add --transport http canonn https://canonn.ai/mcp --header "Authorization: Bearer YOUR_CANONN_API_KEY"

Connect Codex

Keep the key out of your saved configuration by reading it from an environment variable:

export CANONN_API_KEY="YOUR_CANONN_API_KEY"
codex mcp add canonn --url https://canonn.ai/mcp --bearer-token-env-var CANONN_API_KEY

Connect OpenCode

OpenCode reads its MCP servers from opencode.json - in the project, or in ~/.config/opencode/ for every project:

{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "canonn": {
      "type": "remote",
      "url": "https://canonn.ai/mcp",
      "enabled": true,
      "headers": { "Authorization": "Bearer YOUR_CANONN_API_KEY" }
    }
  }
}

Canonn can also be selected as a model. Run /connect, choose Other, enter the provider id canonn and paste the key, then declare the provider:

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "canonn": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "Canonn",
      "options": { "baseURL": "https://api.canonn.ai/v1" },
      "models": { "canonn-r1": { "name": "Canonn R1" } }
    }
  }
}

Canonn R1 answers from your sources and cites them. It is not a general coding model, so keep your agent's own model for writing the code.

Create an API keyRead the API docs