The Aira MCP connects your own AI assistant, such as Claude or ChatGPT, straight to Aira. Once it is connected, the assistant can read and act on your organisation's companies without anyone copy-pasting between Aira and the chat. It is part of Aira Connect. You connect your assistant yourself, and you manage every connected assistant in Settings > AI assistants.
You might see this called the "Aira MCP Server", MCP, or the Model Context Protocol connector. It is a companion to The Aira Connect API: the same Aira data, a different door. The API is for server-to-server integrations; the MCP is for your AI assistant.
Who can use it
The Aira MCP is available to Aira Connect organisations.
It is connected per person, not per organisation. Each colleague connects their own assistant under their own Aira identity, and it acts as that person, in that organisation only. There is no single shared connection for the whole organisation, so everyone who wants an assistant connected sets it up themselves.
Connect your assistant
You add Aira as a custom connector inside your assistant. There is no API key, no client id, and no secret to copy.
In your assistant (Claude, ChatGPT, Claude Code, or another MCP client), add a custom connector and paste this server address:
https://mcp.aira.app/mcpA browser window opens with an Aira consent screen. Approve it.
The connection goes live. Aira appears in your assistant's connector list as "Aira MCP Server".
Choose the right organisation before you approve
The consent screen is where your organisation is chosen, and that choice is effectively one-shot. After you approve once, Aira auto-approves the same assistant every time it reconnects, so the consent screen does not appear again.
If you belong to more than one organisation, check the organisation on the consent screen before you approve. Someone who approves the wrong organisation cannot fix it by reconnecting, because the screen will not come back. The fix is to disconnect that assistant in Settings > AI assistants and connect again.
What your assistant can do
Once connected, your assistant can work with your organisation's companies through a fixed set of 17 tools. It can:
Look up and read company records from Aira's company database, and read the companies, contacts, and registry filings in your portfolio.
Get news for a single company, and news across your whole portfolio.
Read views that have been shared to your organisation.
Add companies to your portfolio and submit deep research.
Two actions spend your organisation's allowance: adding companies to the portfolio, and submitting research. There is no sandbox and no test mode, so every call your assistant makes is live against your real organisation.
Adding a company also writes it to a connected CRM. Removing that company from Aira afterwards does not remove it from the CRM, so clean up on the CRM side if you need to.
Ask about news across your companies
Your assistant can pull news for a single company or across your whole portfolio, so you can ask what has been happening lately without opening Aira. Two things to expect: on a large portfolio it can take a little while to come back, and the first request after a quiet spell may fail once and then work on a retry. Asking what is happening across your companies is fair, it just is not instant.
Two behaviours that look like gaps but are not
Saved views. Your assistant can only see views that have been shared to your organisation. A view you saved just for yourself is not reachable through the MCP at all. So "I have saved views but the assistant says I have none" is expected. To make one available, share it to your organisation.
Registry filings. Filings cover Sweden and Norway only, with Swedish records starting from September 2025. Nothing in the response marks what is and is not covered, so silence does not prove that nothing was filed. For a large or international portfolio, treat an empty filings result as "not covered here", not "all quiet".
Manage connected assistants
You manage connected assistants in Aira on desktop, under Settings > AI assistants. The page has four parts:
Your AI assistants: the assistants you have connected.
Everyone's AI assistants: if you are an owner or admin, every assistant connected to your organisation, grouped by the person who connected it. From here you can set whether an assistant can only read your data or also make changes, disconnect a single assistant, or disconnect everything one person has connected.
Organization settings: two switches that apply to your whole organisation. Allow AI assistants controls whether anyone new can connect. Allow assistants to make changes is the ceiling on what any assistant can do: turn it off and every assistant can only read, and nobody can grant more, even to themselves.
Restore access: someone you disconnect stays disconnected until you restore them here.
The two organisation switches apply when someone connects an assistant, so assistants that are already connected keep working. Turning off Allow AI assistants stops new connections but does not stop the ones already running: disconnect those under Everyone's AI assistants.
One person can have several live assistants at the same time. Reinstalling an assistant, or connecting from a second laptop, registers a new one, so the list can show three entries all called "Claude Code". Disconnecting everything for that person in a single step is the clean way to offboard them.
Settings > AI assistants shows a Last used time for each connection and flags older ones with a "Not used recently" badge. Those flagged connections are usually leftovers from earlier sign-ins and are safe to disconnect, so a list with several entries all called "Claude Code" looks busier than it really is. Tidying them up is housekeeping, not a security step.
If your assistant is refused
A common problem is a refused change. An assistant set to read only still sees all 17 tools, so it will try to add a company or submit research and be refused. The refusal is final: the assistant cannot pause and ask for permission in the middle of a conversation, so it just reports that it could not do it.
To fix it, a person with access opens Settings > AI assistants and gives that assistant permission to make changes. Two things have to be true: the organisation's Allow assistants to make changes switch is on, and that specific assistant is allowed to make changes. Once both are set, the assistant works. Nothing needs reconnecting on the assistant's side for a permission change.
If your assistant suddenly stops working
An assistant that has been idle for a while can quietly stop working. It does not warn you and it does not fix itself. When this happens, reconnect or re-authenticate the assistant in your client and it starts working again. This is one of the most common things people run into, and it does not mean anything is wrong with your account.
Other things worth knowing
Disconnecting in Supabase is not disconnecting. If you find a separate sign-in list in Supabase and remove access there, your Aira access is still live. Aira's own list, in Settings > AI assistants, is the authoritative one. Disconnect there, or you will believe you have offboarded when you have not.
If an admin disconnected you, you cannot reconnect. Once an owner or admin disconnects someone, that person cannot connect again until an admin restores their access under Settings > AI assistants, in Restore access, and the reason is not shown to them. Reconnecting simply fails. If this happens to you, ask an owner or admin in your organisation.
A new Aira tool is not showing up. Your assistant caches the tool list when the connection opens, and reconnecting does not refresh it. After Aira ships a new or changed tool, fully restart your assistant rather than reconnecting, so it fetches the current tools.