Connect your Perfect Wiki knowledgebase to Claude, ChatGPT and other MCP-compatible AI assistants via the Perfect Wiki MCP server. Once connected, your assistant can read, search, create and update pages in your knowledgebase directly from the chat.
The MCP server endpoint is:
https://api.perfectwiki.xyz/mcpHow authentication works
You connect by signing in with your Perfect Wiki account (OAuth). During the connection you choose which knowledgebases to expose and which permissions to grant — read-only or read & write. There are no tokens to copy, and you can revoke access at any time.
Claude (web, desktop and mobile)
- In Claude, open Customize → Connectors
Open the Perfect Wiki connector via the link or find Perfect Wiki and click the + button to connect to Claude
Choose your rights - Read only or Read & Write and the knowledge bases access - all or a selected one
- Click Allow Access
When quering Perfect Wiki for the first time, you'll need to Allow the Access to your content
- After that you can read, update and query your Perfect Wiki content through Claude
ChatGPT (Developer Mode)
- In ChatGPT, enable connector support: Settings → Connectors → Advanced → Developer mode.
- Add a new connector with the server URL https://api.perfectwiki.xyz/mcp and choose OAuth as the authentication method.
- ChatGPT opens the Perfect Wiki sign-in and consent screen — pick your knowledgebases and approve access.
- Enable the connector in a conversation to use the tools.
Claude Code
claude mcp add --transport http -s user perfect-wiki https://api.perfectwiki.xyz/mcpThen run /mcp inside Claude Code and follow the browser sign-in to authorize the connection.
Tip: the -s (--scope) flag controls where the server configuration is stored:
- -s local (default) — saved to the current project only.
- -s user — saved to your user-level Claude config (~/.claude.json), so the server is available across all your projects on this machine. Recommended if you want to configure it once and reuse it everywhere.
- -s project — saved to .mcp.json in the project and checked into git, so the whole team gets the same server.
Tools
Once connected, your assistant gets the following tools. They cover both reading and writing, so you can ask it to look something up and to draft or update a page for you — for example: "Update the onboarding page with these three new steps."
- ask — ask a natural-language question and get an answer with the related pages
- create_page — create a new page
- delete_page — delete a page
- get_knowledgebase — get info about a connected knowledgebase
- get_page — read the full content of a page
- list_knowledgebases — list the knowledgebases the connection can use, with their access level
- list_pages — list all pages in a knowledgebase
- update_page — update an existing page: title, content, tags or parent page
Writing tools (create_page, update_page, delete_page) are only available if you granted Read & Write access when connecting.
Permissions and knowledgebase selection
On the consent screen you grant permissions:
- wiki:read — list pages, read page content, and ask AI-powered questions
- wiki:write — create, update and delete pages
You also choose which knowledgebases the connection can reach — all of them, or a selected one. If the connection covers several knowledgebases, tools take a knowledgebase_id so your assistant knows which one to work in; with a single knowledgebase it can be omitted.
Verifying the connection
Once configured, your client should list the Perfect Wiki tools above. Try asking your assistant: "List the pages in my Perfect Wiki knowledgebase."
FAQ
Can one connection access multiple knowledgebases?
Yes. On the consent screen you can grant access to all your knowledgebases or to a single selected one. When several are available, ask your assistant to work in one by name and it will pick the matching knowledgebase.
Can the assistant delete or overwrite my pages?
Only if the connection has write access, which requires the wiki:write permission granted on the consent screen — connect with read-only permissions if you want a safe setup. Most MCP clients additionally ask for your confirmation before running a write tool.
How do I revoke access?
Remove the connector in your assistant's settings — its access stops working shortly after.
The assistant says a tool needs a permission it doesn't have. What's wrong?
The connection was approved with fewer permissions (e.g. read-only). Remove the connector and reconnect, granting wiki:write on the consent screen.
Is my knowledgebase content sent to the LLM provider?
Yes. When you ask Claude (or another MCP client) a question, the tool results it pulls from Perfect Wiki are passed to the LLM as part of the conversation, following your LLM provider's data policy. The Perfect Wiki MCP server itself only forwards the requests your client makes.
My client doesn't show the Perfect Wiki tools. What's wrong?
Most common causes: the client wasn't restarted after editing the config, the connection wasn't completed (finish the sign-in and consent flow), access was revoked, or the transport type in the config doesn't match http. Check your client's MCP logs for the exact error.
Can I use the server with LLMs other than Claude and ChatGPT?
Yes. Any MCP-compatible client works — Cursor, Windsurf, Zed, VS Code (Copilot agent mode), custom agents built on the MCP spec, etc. The server is model-agnostic.
Is there a request quota or rate limit?
The MCP server is backed by the Perfect Wiki API, so the same quotas and rate limits as your plan's API apply. Heavy usage (e.g. bulk list_pages loops) can hit those limits — prefer the ask tool for broad questions over iterating every page.
Why aren't reading functionalities listed among the Claude capabilities?
Our MCP server sends the same list of 8 tools to every connection, regardless of whether you choose "Read only" or "Read & write" on the consent screen: list_knowledgebases, get_knowledgebase, list_pages, get_page, create_page, update_page, delete_page and ask. We verified this against production — the access level you pick changes what the tools are allowed to do, never which tools Claude is shown. So the read tools are definitely being sent to Claude in your case too.
Why aren't reading features working on Claude?
That can happen if the organization settings for the connector are turned off by default. Admins can manage the settings via Claude directory.
_6d3c8648-a41d-435a-95f8-0019f2a3955b.png)
Need help?
If something doesn't work as expected, contact support.