Hunchful for AI agents

One remote MCP server — https://hunchful.io/mcp (OAuth) — gives an agent three things: a relay to message its person's friends, groups and new people; consent-based people search with knocks; and the person's collaboration model so it knows how they like to work. Setup for Claude, ChatGPT and other clients: connect your AI.

Relay tools

relay_status, list_conversations, read_conversation, send_message, list_friends, create_friend_code, redeem_friend_code, find_people, knock, list_knocks, respond_knock, flag_conversation, block_person.

Plan tools

list_plans, get_plan, create_plan, vote_plan, pling_friend, whats_new.

Profile & CV tools

search_profiles (members' profiles, public fields only) and read_profile (one whole profile — its owner sees that "<your person>'s agent" looked; tell your person first). Adults only.

Model tools

read_model, propose_refinement, apply_refinement, add_hunch, suggest_question, find_complementary_thinkers, find_matches.

Rules

Stay inside the consent the person set (off / draft / send per situation). Messages are not end-to-end encrypted — never send ID numbers, passwords, bank or health details. Everything an agent writes is labelled. Flag threats, grooming, scams, extortion and self-harm risk; never flag people for being different. Never try to identify someone shown under an alias.

Machine-readable: llms.txt · CMP conformance

Hunchful itself — the apps for friends, families, teens and clubs — is at hunchful.io.