Connectors
Vultur for assistants
Connect Claude, ChatGPT or anything else that speaks MCP to your Vultur account, and ask it who you know. It searches your own contacts and the networks of friends who have shared theirs with you — the same people you see on the Search page, through the same boundaries.
What it is
Vultur publishes an MCP server: a standard way for an assistant to use a tool on your behalf. You add it once, sign in, and from then on your assistant can search your network inside whatever conversation you are already having.
https://app.vultur.ai/api/mcp
You sign in through the normal Vultur login, in a browser, and approve the connection there. There is no API key to create, copy or keep safe. You can disconnect it from your assistant at any time.
What an assistant can see
- Your own network. The contacts Vultur has already built from the accounts you connected: names, headlines, job titles, companies, locations, LinkedIn URLs.
- Your friends’ networks, at arm’s length. If a friend on Vultur has shared their network with you, people in it appear in your searches with that friend’s name attached, so you know who to ask for the introduction.
- Which friends are currently in scope, and who in your contacts is worth adding as a friend.
What it cannot see, and cannot do
- Nothing of a friend’s beyond their name. When a person is reached through a friend, the only thing that crosses the friendship boundary is that friend’s display name. Never their messages, never their contact details, never how often or how recently they spoke to anyone.
- Nothing outside your Vultur account. It is the same scope as your own Search page. An assistant cannot widen it, and removing a friendship closes that network immediately.
- It reads; it never acts. There is no tool that sends a message, makes an introduction, adds a friend or changes anything. If your assistant offers to email someone it found, that is its own idea, not something Vultur gave it.
Results describe real people and are treated as information, never as instructions.
The nine tools
Your assistant picks these itself. They are listed so you can tell what it did.
- get_usage_instructions
- How to search well, and what the results mean.
- search_network
- Finds people from a plain-language description, across your contacts and your friends' networks. Returns up to 25 at a time, in relevance order, without scores.
- check_candidates
- Reads up to 30 of those profiles properly and scores each one against what you asked for, with a quote as evidence.
- refine_search
- Narrows a search you already ran — “only the ones who have raised a Series A” — without starting over.
- next_page
- More results from the same search.
- get_person
- One person's profile, if they are still reachable in your network.
- list_friends
- Which friends you have on Vultur, and whose networks are currently included in your searches.
- suggest_friends
- People in your own network worth adding as friends. Names and reasons only — no contact details leave Vultur.
- get_usage
- How much of today's allowance is left.
Searching and scoring are separate steps. A search comes back broad and without scores, and your assistant then picks a handful of people to read properly. So “23 results” is not 23 matches — it is 23 people worth looking at.
A search also reports anything it could not check. If you ask for “people who are open to work” and Vultur has no way to know that, it says so, and a well-behaved assistant repeats that back to you instead of implying the filter was applied. If yours does not mention it, ask.
How to connect
- Open Connectors in Vultur and pick your app. It walks you through the rest.
- Anything else: add the endpoint above as a remote MCP server, sometimes called a custom connector. Your assistant handles the sign-in.
- If your assistant has no way to add a server by URL but can read a link, hand it
https://app.vultur.ai/agents/llms.txtand ask it to connect itself. Those pages are written for assistants rather than people.
On Claude Team and Enterprise plans an owner has to add the connector for the organisation before anyone can use it. The Connectors page has a link to send them.
Your daily allowance
An account can make about 1,000 calls a day, counted from midnight UTC. That is a brake on an assistant stuck in a loop, not a ration: an ordinary question costs a handful of calls, and a long afternoon of research will not come close. If you hit it, your assistant is told how long to wait. Talk to us if you have a reason to need more.
Connectors shows what your assistants have been doing — what ran, when, and whether it worked.
When it does not work
- “Failed to connect” straight away. Usually the URL. It ends in
/api/mcp, with no trailing slash. Check you copied the whole thing. - It asks you to sign in again and again. The approval did not stick. Remove the connector in your assistant, add it back, and complete the browser sign-in without closing the window early.
- It connects but finds nobody. Check your network has finished building on the Network page, and ask your assistant which friends are in scope — a search with no friends in it is a much smaller search.
- It says the daily limit is reached. Something looped. Wait for the reset, and look at Connectors to see what ran.
- It claims a filter it was told it could not apply. Ask it what the search reported as unsupported, then judge the answer again.
Still stuck, or something looks wrong with what it can see? Open Help & feedback in the app sidebar and a human will answer.