How to give Claude access to your team wiki (MCP setup in 5 minutes)

Your team's knowledge lives in a wiki. Your AI assistant needs that knowledge. The clean way to connect the two is the Model Context Protocol (MCP) — and if your wiki is a tela wiki, wiring up Claude takes about five minutes and no glue code. This is the practical, step-by-step version. (For the concept — why a wiki being an MCP server matters — start with What is an MCP wiki server? first.)

What you'll end up with

Claude able to search, read, and write your wiki directly — over a typed protocol, with your permissions, no copy-paste and no browser automation. Ask "what's our on-call runbook?" and Claude retrieves the actual page with citations; ask it to "draft a postmortem in the incidents space" and it writes one back.

Before you start

You need a tela wiki. Either:

Cloud — a free workspace at telawiki.com, or
Self-hosted — tela is open-source (AGPL-3.0); run it with Docker on your own box.

Either way your MCP endpoint is your wiki's URL + /api/mcp (on cloud: https://telawiki.com/api/mcp).

Step 1 — Connect Claude over OAuth (the modern path)

Modern hosts speak MCP over HTTP and handle auth for you:

In Claude (Desktop or Code), add a custom connector.
Paste the endpoint: https://telawiki.com/api/mcp (or your self-hosted URL).
When the tela login screen appears, sign in and consent. That's it — no token to copy.

Claude discovers tela's tools at runtime. You'll see search, research, get_page, create_page, update_page, and the rest appear as callable tools.

[!TIP] On Cursor or VS Code, skip the manual step — the MCP setup page has one-click install buttons that pre-fill the endpoint for you.

Step 2 — Older host? Use the npm proxy

If your host only speaks stdio (older MCP clients), bridge it with the published proxy:

Create a Personal Access Token in tela (Settings → API keys), scoped read or read-write.
Point the client at the tela-mcp npm package, passing the token as TELA_API_KEY and your wiki as TELA_BASE_URL.

The proxy is a dumb stdio↔HTTP pipe to the same endpoint, so the tool surface is identical.

Step 3 — Try it

Ask Claude something only your wiki knows:

"Search the wiki for our deployment checklist and summarize it." → it calls search/research and answers with citations.
"Create a page called 'Q3 retro' in the Team space with these notes." → it calls create_page and hands you back the link.

Scopes and safety

Two things worth knowing before you let an agent write:

Tool calls run with your account's permissions. An agent can only touch what you can — nothing more.
Read and write are separate scopes. A read-only token or connection never mutates anything. Grant write only when you want the agent authoring.

More on the trust model in Giving AI agents safe access to your wiki.

Why tela specifically

Plenty of wikis now ship an MCP server. tela's difference is ownership: your pages are canonical markdown you keep, it's self-hostable and open-source, and the MCP server is built in — not a paid add-on. If you're weighing options, see how it stacks up against Notion, Confluence, Docmost and other wikis.

Give Claude a wiki it can actually read and write, and "ask the docs" stops being a copy-paste chore.