> ## Documentation Index
> Fetch the complete documentation index at: https://crsdk.app/llms.txt
> Use this file to discover all available pages before exploring further.

# Troubleshooting

> Common issues and solutions when using the MCP servers

While we've taken steps to ensure the smoothest experience possible, you may encounter the following issues in the beta release.

### "Could not connect to MCP Server \[server name]"

<Check>
  **Most likely issue:** Connection to the server timed out.
</Check>

**Solutions:**

* Check to make sure your configuration aligns with the setup instructions
* Stop generating the current response, refresh the page and try again
* Disconnect from and reconnect to the connector
* If neither of these work, delete the connector and add it again
* If it still doesn't work, contact [jordan.lee@sony.com](mailto:jordan.lee@sony.com) and [report a bug](https://sna-ipsa-developer-tools.atlassian.net/jira/software/form/68c76742-1683-4ac8-83f3-a2e2518e5ffc?atlOrigin=eyJpIjoiMGNmZDVmZmE4MGQyNGZiNGFlZTQ1MGVhYTU2M2JkODciLCJwIjoiaiJ9).

### Claude makes 10+ tool calls before providing an answer.

<Check>
  **Most likely issue:** Claude is doing an exhaustive search of the help guides for completeness.
</Check>

**Solutions:**

* Use the [system prompt](/mcp-server/system-prompt) to refine Claude's search approach.
* Modify the system prompt to instruct Claude to prioritize quick answers over deep research.
* Interrupt Claude's response and ask it to summarize the information it found.

### "Tool \[tool name] not found"

<Check>
  **Most likely issue:** Connection to the server was lost. Tools not available for use.
</Check>

**Solutions:**

* Disconnect from and reconnect to the connector
* If this doesn't work, delete the connector and add it again

### "Claude's response was interrupted"

<Check>
  **Most likely issue:** Connection to the server is successful, but connection was temporarily lost due to startup not complete.
</Check>

**Solutions:**

* Refresh the page, wait 30 seconds before making a query
* Make your first request smaller. After successful response, resume normal operation

### "Tool call \[tool name] returned no results or 429 error"

<Check>
  **Most likely issue:** Rate limit exceeded (20 requests/second or 100 requests/minute)
</Check>

**Solutions:**

* Wait several seconds before making another request
* If your LLM is making rapid consecutive requests, prompt it to space out tool calls
* If the tool call body has a `top_k` (number of content to retrieve) greater than 5, prompt it to keep `top_k < 5`:

  ```json theme={null}
  {
    "query": "exposure compensation",
    //top_k indicates the number of results to return in the response
    "top_k": 1
  }
  ```