Coalex.ai — AI Governance Platform | Human-in-the-Loop

• Is this agent on Coalex? — Call this first. Is the agent in this repository already on Coalex? Before calling, look in the repository (do not send any file contents) and report only what you found: - agent_id: the id the agent uses, e.g. in `coalex.coalex_context(agent_id=...)`, or the agent's service name if there is none - sdk_dependency: `coalex` (Python) or `@coalex-ai/sdk` (TypeScript) is a declared dependency - register_call: the code calls `coalex.register(` - auto_instrument_call: the code calls `coalex.auto_in
• Integration guide — Step-by-step instructions to integrate this repository's agent with Coalex and run it once. language: the agent's language, e.g. 'python'. framework: the agent framework. Pass the one that matches the repository; the frameworks that have a guide are python: 'foundry', 'langgraph', 'openshell'. Omit it, or name another, for the generic guide. Follow the steps in the developer's code. The guide ends with finding how the agent is started, running it once, and writing that command into the agent's
• Check the Coalex SDK — Is the Coalex SDK in this repository current? language: 'python' (package `coalex`) or 'typescript' (package `@coalex-ai/sdk`). installed_version: the version locked or installed in the repository (from uv.lock, poetry.lock, requirements.txt, package-lock.json...), or omit it if the SDK is not installed. Reads the latest release from PyPI or npm.
• Connect a Coalex account — Connect this MCP client to a Coalex account. Call it when a result's next_step names it. Without an account, calling it makes the MCP client open the Coalex login in the browser; the developer logs in once. Then it returns the organisation, and the tools that read the organisation's data (agents, traces, escalations, prompts, compliance) are listed. If they do not appear, the client has to reconnect (in Claude Code: /mcp).

Schemas and invoke metadata: Directory API