Connect Claude, ChatGPT and other AI assistants
Connect an AI assistant to MetaOpus, what it can look up for you, and what it never sees.
Open AI assistantsMetaOpus works as a connector for AI assistants that support the Model Context Protocol (MCP) — Claude, ChatGPT, Gemini, Microsoft Copilot and others. Once connected, you can ask your assistant about requirements, your bench and your submissions in plain words, and it looks the answers up in MetaOpus as you.
Before you start
- Your firm's administrator turns AI assistants on under Settings → AI assistants, and chooses which roles may connect.
- You need your own MetaOpus login. The assistant signs in as you and never sees more than you can see in MetaOpus.
- Your AI assistants page shows the address to use, and says so if your firm hasn't turned it on for your role.
Connect
- Copy the addresshttps://mcp.metaopus.ai/mcp — also shown on your AI assistants page.
- Add it to your assistantClaude: Settings → Connectors → Add custom connector. ChatGPT: Settings → Connectors (developer mode). Paste the address.
- Sign in to MetaOpusYour browser opens MetaOpus. Sign in as usual, including multi-factor authentication if your account uses it.
- ApproveThe consent screen names the assistant, your role, your firm, the tools and what they can return. Approve to finish.
What you can ask
- Recruiters: find requirements by skill, place, engagement and date; open one in full; see who on your bench fits it; what's out there for a consultant; where your submissions stand; what the market is asking for.
- Leads, managers and directors: your team's desk, compared by recruiter or by branch; who is away and who covers; what is overdue.
- Consultants, where the firm has included them: your matches, your submissions, your documents and your profile.
What it never sees
- Where MetaOpus's requirements come from.
- Contact details — email addresses, phone numbers and apply links are withheld, including from description text. Open the requirement in MetaOpus to see them.
- Anything your role can't see in MetaOpus, rates for consultants, and any firm's margins.
Every tool is read-only: an assistant can't change, send or spend anything in MetaOpus, and a connection never spends a credit.
Ready-made questions
In Claude, the + menu lists MetaOpus questions you can start from — your morning brief, C2C roles for a skill, who on your bench fits a requirement, your team's week, what the market wants. You only see the ones your role can use.
Disconnect
Your AI assistants page lists every assistant you have connected, when it was last used, and a Disconnect button. You can also pause assistants for yourself, which disconnects them until you turn it back on. Your administrator can end any connection at the firm, turn it off for one person, or turn it off for everyone; each takes effect at the assistant's next request.
Frequently asked
Does MetaOpus see my conversation with the assistant?
No. The assistant sends MetaOpus only the question it needs answered — a search, an id — and MetaOpus records which tool was used, when and how many rows went back, never the conversation.
Where does the answer go after MetaOpus sends it?
To your AI assistant, under your own account with that provider and its terms. MetaOpus does not control how long the provider keeps it, which is why contact details and consultants' personal details are withheld by default.
Why was my request refused?
The assistant was asked for something your role can't see, or your firm hasn't enabled. The refusal says why in plain words and what you can do instead.
Who can see what my assistant looked up?
Your firm's administrators see every AI assistant call in Settings → AI assistants → Activity, as they see other activity in the audit log.
Was this helpful?
More in Account & security
- Account details, billing and your team What an invoice needs, which documents are held, and how people are added.
- Roles, sign-in security and the audit log Who can see what, how sign-ins are protected, and what is recorded.
- Why a setting is what it is Every level that was consulted, what is never a setting, and the changes that need two people.
- Where the work comes from Sources, what each is contributing, and why a healthy one can sit at zero.
- Mail, storage and model connections What healthy actually means, and why a test sends a real message.
Still stuck? Email us what you are trying to do — a person answers, and we would rather fix the product than explain it twice.