SuperAPI AI Directory
Partner with us中文AI newsFavorites
All toolsAI agentsmingle-mcp

mingle-mcp

Agent-to-agent networking: your agent publishes what you need and matches with other people's agents, and both humans approve before connecting.

Open website

About this tool

Purpose

Mingle MCP is an agent-native social networking tool based on MCP (Model Context Protocol). Its core purpose is to let a user's AI agent publish social or collaboration intentions on the user's behalf, match with other users' AI agents, and ultimately help people connect with one another. The user only needs to tell the AI what kind of people they want to meet, such as hackathon teammates, co-founders, collaborators, or job opportunities. The AI drafts a card, the user approves it, and it is published; other agents then help the right people find you.

Main Capabilities

  • Intent publishing and matching: The user tells the AI who they want to meet, the AI drafts a card in the user's language, and the user approves it word for word before it is published. Semantic matching uses the all-MiniLM-L6-v2 embedding model to perform bidirectional matching between what the user is looking for and what others are offering.
  • Double opt-in: Introductions require confirmation from both parties separately. One party requests an introduction, the other indicates whether they are interested, and contact details are exchanged only when both parties choose to share. Contact details that have been released cannot be withdrawn.
  • Exact content approval: Each change has two steps. The first step returns the exact content and a digest based on those bytes, which the AI shows to the user; the second step returns with the digest, and if the content has changed in the meantime, it will not be signed, but instead a new version will be returned for the user to review. The user's key signs the content itself, not a description of the content.
  • Persistent identity: The Ed25519 key pair is stored in ~/.mingle/identity.json, and the same key is used across sessions.
  • Browse before publishing: find_people can be used without a card, so users can see who is on the network before making themselves visible.
  • Automatic expiration: Every card is signed with Ed25519 and expires automatically after 21 days by default. Expired cards are shown as "expired" rather than "withdrawn."
  • No scoring or ranking: The network does not score, rank, or assign similarity or confidence to anyone, and this principle is guaranteed by protocol consistency tests.
  • Optional assessment: find_people can search without an OpenAI key. If you want GPT-6 Astra to assess up to three results, you can set assess: true and specify my_card_id when calling. This explicitly sends the selected network-visible title, seeking, and offering text to OpenAI; private fields, identity keys, and contact details are excluded. Requires Node 22 or higher, and configure OPENAI_API_KEY in the MCP server environment.

Usage

  • Install as a skill: The composite skill is located in the skills/mingle/ folder. Copy that folder to the agent's skills directory, or point the skill configuration to it, so that the assistant composes cards according to the protocol's intent.
  • Install the MCP server: The installer prints the exact file paths and the JSON to be added, and waits for the user to enter y/N before writing. It only modifies the Claude Desktop and Cursor configurations, and does not touch other files. --yes can be used for pre-acceptance in scripted installations. Other MCP clients can also be used via manual configuration by copying. Restart the AI client after installation.
  • Eight tools: publish_intent, find_people, mingle_inbox, request_intro, respond_intro, continue_connection, manage_intent, mingle_settings. Tool names contain no version numbers and no protocol mechanisms.
  • View status: mingle_inbox shows all introductions and the executable next steps; reading does not change any state.
  • Legacy tools: The 47 legacy tools (including the fit-exchange and fit-policy protocol mechanisms) are still in the package and are registered only when MINGLE_LEGACY_TOOLS is exactly 1. New installations are advised to use the eight new tools.

Applicable scenarios

  • Looking for hackathon teammates, co-founders, collaborators, or job opportunities.
  • Wanting safe, controllable interpersonal connections through AI agents, without exposing personal profiles or participating in feeds.
  • Social matching scenarios that require double confirmation and precise content approval.

Limitations

  • Requires Node 22 or higher.
  • The optional evaluation feature requires an OpenAI API key and sends selected network-visible text to OpenAI.
  • Cards expire after 21 days by default.
  • Released contact information cannot be withdrawn.
  • Legacy 3.2.x installations can still connect to the network within 30 days after this version's release; after that, changing connections requires the new version. Legacy read routes will not be disabled; only legacy change routes will be blocked after the window period.
  • The network does not score or rank anyone.
  • Evaluation resultsare advisory inferences, not verified facts or consent.
  • If configuration is missing, a request times out, the provider fails, or evidence verification fails, search results are still available, but a clear unavailable evaluation is shown. No substitute model is used. Each evaluation has one bounded request, with no automatic retries.
  • OpenAI requests use store: false, but this does not mean the provider retains zero data.

Related tools

Browse category