Skip to main content

​
Connection problems

  • Make sure the URL is exactly https://api.evermuse.com/api/mcp, with no trailing slash and no /sse suffix.
  • Make sure your client supports remote MCP servers with OAuth. Older clients that only support local (stdio) servers can’t connect directly.
  • Leave any OAuth client id or secret fields empty: Evermuse registers clients automatically.
  • If you clicked Deny, or the sign-in timed out (authorization requests expire after 10 minutes), start the connection again from your client.
The access token expired or was revoked. Most clients refresh automatically. If yours doesn’t, disconnect and reconnect Evermuse. Tokens are also revoked when you’re removed from the workspace, and a refresh token that is reused after it has rotated revokes the whole connection. In both cases, reconnect.
Each connection is bound to the workspace you chose at sign-in. Disconnect Evermuse in your client, reconnect, and choose the right workspace on the consent screen.
Evermuse uses stateless Streamable HTTP with JSON responses. Configure your client for Streamable HTTP (often labelled http), not the legacy SSE transport.
Custom clients must send Content-Type: application/json and Accept: application/json, text/event-stream on every POST.

​
Tool problems

Evermuse data is organized by product. Ask the agent to call get_products and use the right product_id. If the workspace is new, make sure you’ve connected sources or uploaded meetings and documents in the Evermuse app first.
The tool needs a permission the connection doesn’t have. Signal writes (add_signals, update_signals) need mcp:write. Reconnect from a client that requests the default scopes, or create an API key with mcp:write.
The workspace has used its included credits and has no active plan. Upgrade in the Evermuse app or contact team@evermuse.com. find_skills and read_skills keep working.
add_source queues processing. It doesn’t wait for it to finish. Use find_sources (browse mode lists sources still processing) to check its processing_status, and search again once it’s done.
Each access token is limited to 1,000 requests per hour. Wait the number of seconds in retry_after. Agents that loop over many pages can pass include_guidance: false after the first call to get more data per request.
Search-shaped tools trim pages to fit client limits and return next_offset. Ask the agent to continue from it, or to narrow the search with filters such as project_id, source_types or a date range.

​
Local-storage workspaces

Workspaces set up with Evermuse Desktop to keep data on your computer don’t store research data in the cloud. For those workspaces, the cloud MCP server serves only find_skills and read_skills. Other tools return Tool "…" is not available for this workspace. Use the Evermuse Desktop app’s MCP connection for those workspaces.

​
FAQ

No. Any Evermuse account can connect, and new workspaces include free credits. Each successful tool call uses one credit, and the skills tools are free. See Credits.
Only through the write tools (five on a standard workspace). The agent can add sources and signals, edit signals, and create or edit shaping notes. No tool deletes sources, signals, notes, products or projects. Clients can ask you before running any tool that isn’t read-only, and shaping-note edits can be reverted from the note’s version history.
No. Evermuse only receives the tool calls your client makes and their arguments. It doesn’t receive the rest of the conversation.
Each connection covers one workspace. To work in another workspace, disconnect and reconnect, choosing the other workspace on the consent screen.
No. Each person connects with their own Evermuse account, so access checks and attribution stay per user.
Remove Evermuse from your AI client, or see Revoking access for every option.

​
Get help

Email team@evermuse.com with the name of your AI client, the time of the problem and the error message. Report security issues to security@evermuse.com.