Agents

LLM and MCP surface

CommentHunter publishes the same capability contract in three machine-readable forms so agents do not have to scrape HTML.

• llms.txt — short index of endpoints, constraints, and docs links (llmstxt.org convention).
• llms-full.txt — expanded operation list for long-context agents.
• openapi.json — OpenAPI 3.1 for tool calling and SDK generation.

How to connect an agent today

1. Fetch https://commenthunter.click/llms.txt at session start.
2. Import https://commenthunter.click/openapi.json as tools (ChatGPT Actions, Claude connectors, Cursor MCP wrappers).
3. Call extract_comments with a public URL. Confirm paid checkout with the user before create_checkout.

MCP

A hosted Streamable HTTP MCP server (OAuth 2.1 + PKCE, ChatGPT-style) is on the roadmap. Until it ships, wrap the OpenAPI document with your MCP runtime (for example an OpenAPI-to-MCP proxy) pointing at the Supabase functions base URL. Tool names should match OpenAPI operationId values.

Safety for agents

• Only public URLs the user provided. Do not attempt credential stuffing or private-account access.
• Destructive or paid operations (create_checkout) require explicit user confirmation.
• Do not log JWTs. Store them in a secret manager.
• Respect rate limits and credit balances from check_subscription.