Guide
Querying data with Fibery MCP is slow. How to improve quality/speed of answers?
Make use of skills - in a single session, get LLM to do what you wanted with Fibery MCP. In the end, ask it to pack it into a reusable document/skill so that it can re-used later.
Give as much information as you can - what databases to use, what fields to query, with what filters. All of that will help LLM to really focus.
Lastly, you may copy contents of our Entity API guide as markdown and ask LLM to find examples of how fibery queries are constructed. It will NOT work with all tools, but in case of query, it will, because the input structure of the tool is very close to raw API call.
I see a "Custom MCP server introduces risk" warning
This is a generic warning shown for all MCP servers by the AI client — it does not indicate a vulnerability in the Fibery MCP Server. The Fibery MCP Server follows the same security model as other Fibery integration endpoints.
I see "This MCP server cannot be used in Deep Research/Agent"
This is expected. Fibery's MCP Server does not implement the search and fetch tools that Deep Research and Agent modes require, because they are not relevant to typical Fibery use cases.
Fibery's MCP Server works well for:
Data querying and reporting
Entity creation and updates
Workspace automation and integrations
I see the error "no transport found for session id"
This error means the client cannot find an active session connection. Try these steps in order:
Re-add the connector. Delete the Fibery MCP entry from your client's connector list and add it again. This clears any stale session cache.
Check the client version. Make sure your AI client is up to date and compatible with the MCP server.
Restart the client. A WebSocket session may fail to initialize on first connect. Restarting often resolves it.
If none of these steps help, the issue may be server-side. Contact Fibery support with your session details.
I'm getting a 401 Unauthorized error using the MCP — is this caused by IP restriction?
Yes, most likely. Fibery's MCP server calls the API using your own user token (not a service-to-service token), so if your workspace has IP restriction enabled, requests are blocked unless they come from an allowed IP — even for admins. The IP that matters is the one your MCP client (e.g., Claude) actually sends requests from, not your local machine, and there's currently no way to allowlist a specific MCP client's IPs separately, since they're not guaranteed to stay stable. If this is blocking you, contact us in chat.
How does MCP works with the Rich text?
MCP can update small parts of documents and rich text fields.
Some things are not supported (so far):
MCP Server can also add comments to rich text. You may use it to ask AI review the doc and add comments to some places so you can review them and correct things faster.
Can I connect more than one Fibery workspace to Claude?
Yes, but there's no native "multiple workspace" toggle in the connector itself - you need one small trick. Add /?workspace=[WORKSPACENAME] to the end of each MCP connector URL before adding it in Claude. The parameter doesn't do anything on Fibery's side; it just makes the URL unique enough that Claude registers it as a separate connector instead of merging it with one you've already added.
I connected ChatGPT to Fibery through its Plugins page. How do I make it ask before every change again?
The approval setting lives entirely in ChatGPT application. To change it, go to Settings -> Plugins -> Fibery -> Permissions, and select Always Ask. If that doesn’t work, reconnect Fibery Plugin and check Permissions tab there again.
I'm using ChatGPT Atlas and it changed things on the Fibery site without asking - how do I stop that?
Atlas is OpenAI's own browser, and its Agent mode drives the Fibery web app directly (clicking, typing, submitting) using your logged-in session, the same way you would by hand. By default, Agent mode only pauses to confirm on sites OpenAI treats as sensitive, like banks - Fibery isn't one of those, so it won't stop and ask on its own. Two things help: switch Agent mode to logged-out browsing, which forces it to ask before it signs in as you at all, and add a custom instruction telling it to pause for your approval before submitting or changing anything. Neither is a guarantee - unfortunately it depends on Atlas honoring the instruction, since Fibery has no control over Agent mode's behavior. One more note to be aware of: because Atlas acts through your own logged-in session, Fibery's Activity Log can't tell its edits apart from ones you made by hand - there's no separate actor to filter by, the way there is for the MCP connector.