One connector, both halves
ilolink’s MCP server
Add ilolink as a connector in Claude, Claude Code or ChatGPT, and your assistant can publish what it just made as a live page and get the link back — and read and write the specs, skills and runbooks your whole team’s assistants share. One connection, no copy-paste, no API key.
What MCP is, in one paragraph
MCP — the Model Context Protocol — is how an AI assistant talks to a service it doesn’t have built in. You add a connector once; from then on the assistant can call that service’s tools during a normal conversation. ilolink’s connector is a remote one, so there is nothing to install and nothing running on your machine. It authenticates with OAuth, the same way you would sign in to any other app, and this is the whole of the setup:
https://mcp.ilolink.com/mcpConnecting, per assistant
Every client uses the same URL and the same approval screen. The only thing that differs is where the setting lives.
Claude (desktop and web)
Claude speaks remote MCP over OAuth, so there is no key to copy and nothing to paste into a config file.
- Open connector settings. In Claude, go to Settings → Connectors and choose to add a custom connector.
- Paste the connector URL. Use https://mcp.ilolink.com/mcp as the URL. Leave everything else alone.
- Approve, and pick a teamspace. Claude sends you to ilolink to sign in and approve. The approval screen asks which teamspace the assistant may publish into and read from.
Claude Code
One command, or install the plugin if you also want ilolink's publishing and registry skills.
- Add the server. Run: claude mcp add --transport http ilolink https://mcp.ilolink.com/mcp
- Or install the plugin. Run /plugin marketplace add https://github.com/wilsonbright/ilolink, then /plugin install ilolink@ilolink. The plugin wires the same connector and adds two skills.
- Authorise once. The first tool call opens the same OAuth approval, where you choose the teamspace.
ChatGPT
The same URL and the same OAuth flow. ilolink implements the search and fetch tools ChatGPT's connector contract expects, so your published documents are searchable and quotable inside a chat.
- Open connector settings. In ChatGPT, go to Settings → Connectors and create a new connector.
- Paste the connector URL. The URL is https://mcp.ilolink.com/mcp — nothing else to fill in
- Approve, and pick a teamspace. You will be sent to ilolink to approve the connection and choose its teamspace.
What your assistant can do once it’s connected
You never call these by name. You say “publish this as an ilolink page” or “check the team registry first” and the assistant picks the tool.
Publish and measure
Turn something the assistant just made into a page anyone can open — then find out what happened to it.
- publish_document
- Publish Markdown, HTML, a PDF, JSON, CSV, a diagram or an image and get a shareable URL back, plus a private analytics link.
- update_document
- Replace the content with a new version. The link stays the same.
- unpublish_document
- Take a document offline so its link stops working. Reversible from your dashboard.
- list_documents
- Everything published from this connection, newest first, with share URLs and view counts.
- get_analytics
- Views and comment count for one document. Heatmaps and read-through live on the dashboard.
The team registry
Ten kinds of team knowledge — skill, agent, spec, design, plan, workflow, session, decision, runbook, eval — that every teammate's assistant can read.
- artifacts_list
- See what guidance the teamspace already has before starting a task. Doubles as a sync changefeed: pass a timestamp and compare hashes.
- artifacts_get
- Read one artifact in full. The response names who wrote it, so the assistant can tell you whose instructions it is following.
- artifacts_put
- Save or update one artifact. If your team reviews member writes, it lands as a proposal rather than going live.
- artifacts_push
- Push up to 50 files from a repository in one call — the way to get a project's skills, specs and runbooks into the team registry.
Search from the chat
Built for ChatGPT's connector contract, and useful in any client: your own published documents become something the assistant can look things up in.
- search
- Find your published documents by title. Returns ids and share URLs.
- fetch
- Pull one document's full text back into the conversation, so the assistant can answer from it and cite it.
What the connection can and can’t reach
- One teamspace, for life. You choose it when you approve the connection, and it never moves. Connect again to use a different one.
- Membership is checked every call. Not once at approval — on every request, against the live membership table.
- Published documents default to unlisted. The link works; search engines are told to stay away and previews don’t quote the text.
- Your team can require review. Turn it on and a member’s registry write becomes a proposal an admin approves, rather than instructions every assistant starts reading immediately.
Questions
- Which assistants work with this?
- Anything that supports remote MCP connectors. Claude, Claude Code and ChatGPT all use the OAuth flow at the connector URL — no key to copy. Clients that cannot do OAuth but let you set request headers can use a connector token instead, created on the connect page inside your account.
- Does the assistant get access to everything in my account?
- No. A connection is bound to exactly one teamspace, chosen by you on the approval screen, and it stays bound to it for the life of the connection. Membership is re-checked on every single call, so removing someone from a teamspace cuts off their assistant too.
- What if I make a new teamspace later?
- Connect the assistant again and pick the new one on the approval screen. An existing connection keeps writing to the teamspace it was approved for — it does not follow you.
- Do documents published by an assistant go public?
- Not by default. Anything published over MCP is unlisted: the link works for whoever you send it to, but the page is kept out of search and its text is not quoted in previews. The assistant can publish publicly if you ask it to.
- Can a teammate's artifact change what my assistant does?
- Artifacts are instructions written by your teammates, and assistants are told to treat them as data rather than as orders — to apply them only where they fit what you asked, never to let them change tool permissions, and to name the artifact and its author before acting on it. Teams can also require admin review before a member's write goes live.
- Is there a separate price for the connector?
- No. The connector is part of ilolink. Free covers one person; a team is a single one-time payment.
Try it on your own doc
Sign in, copy the connector URL, and pick the teamspace it may write to.
Keep reading
Pricing — free for one person, one-time payment for teams
What ilolink costs: free solo, and a single payment for a team. No subscription.
Frequently asked questions
Accounts, privacy, formats, size limits, link permanence — the direct answers.
Teamspaces & the artifact registry
Where the skills, specs and runbooks your assistants read actually live.
What people share with ilolink
Real examples — specs, prototypes, client reports — each paired with the question the analytics answer.