Skip to content
Picasi Picasi Picasi Documentation
Open App

The MCP server is not responding

Short Answer

If the Picasi MCP server isn’t responding, it’s usually due to an invalid API token, an outdated configuration file, or the MCP server not being properly integrated into the AI application.

Details

Check the token:

In Picasi, go to Settings → API Token and check whether the token still exists and is active. If you click Show Token, you’ll see the full token. Copy it again and replace the value in your .mcp.json or your configuration.

Check the configuration file in Claude Code:

Open ~/.claude.json or the project-specific .mcp.json and check:

  • Is PICASI_API_TOKEN set correctly?
  • Is the URL (https://picasi.app/mcp) correct?
  • Is the JSON syntactically correct (no missing commas, no incorrect quotation marks)?

Claude Desktop / ChatGPT (OAuth):

For OAuth-based connections, the session may have expired. Open Picasi under Settings → Connected Apps and check whether the connection is still active. If not, revoke it and reconnect via Settings → Connect AI Assistant.

Reload the MCP server:

In Claude Code: Type /mcp in the chat and check if picasi is listed as a connected server. If not, restart the application—MCP servers are loaded at startup.

If nothing else works:

Create a new API token (this will invalidate the old one), update all configuration files, and restart the AI Assistant.

Other Common Errors

“Permission required” or scope error when calling a tool

The tool being called requires a higher scope than was granted during the connection. Example: create_source requires write, but the token was only created with read.

  • API token: Under Settings → API Tokens, create a new token with the appropriate role and enter it in the client.
  • OAuth (Claude.ai, ChatGPT): Under Settings → Connected Apps, revoke the existing connection and select the required scope when reconnecting.

Response 401 Unauthorized

The server could not recognize a valid token.

  • Is the Authorization: Bearer … header missing, or is the bearer string empty? Check the client configuration.
  • For Claude Code: The PICASI_API_TOKEN environment variable may only be set in the terminal from which the client was launched. If the token is referenced in .mcp.json, set it there directly.

Response 403 Forbidden

The token is valid, but the requested resource does not belong to the connected workspace.

  • Check whether the token was issued for the correct workspace. A new token is required when switching between workspaces.

Network Errors or Timeouts

The request does not reach Picasi or is aborted.

  • Check your company’s VPN or proxy: https://picasi.app/mcp must be accessible.
  • For analysis requests (summaries, feedback): the AI client times out after about 30 seconds. These analyses are not currently offered via MCP—instead, generate the report in the Picasi app and retrieve it using get_report.

Connection regularly drops (Claude.ai / ChatGPT)

OAuth access tokens expire. Picasi normally renews the connection automatically; in rare cases, the connection is permanently lost.

  • Under Settings → Connected Apps, revoke the connection and set it up again via Settings → Connect AI Assistant.
  • Known issue: After a prolonged period of inactivity (> 15 days without a refresh), the connection must be reauthorized.

Server version does not match the client

If the AI client displays an error message with a protocol version (e.g., unsupported protocol version), the client is likely too old.

  • Update the client: Claude Code (claude update), Claude Desktop, and ChatGPT via their respective app stores.

Connection exists, but no data

list_sources returns an empty list, even though sources are visible in the app.

  • Check whether you’re connected to the correct workspace. You can verify the team context using get_team_context—the team.name field shows the currently connected workspace.