fix(copilot): prioritize block discovery over MCP and sanitize HTML errors

autogpt_platform/backend/backend/copilot/prompting.py

@@ -11,6 +11,13 @@
 # Shared technical notes that apply to both SDK and baseline modes
 _SHARED_TOOL_NOTES = """\
 
+### Tool priority: blocks before MCP
+When a user asks for an integration (Google Sheets, Gmail, Slack, etc.), always
+search for platform blocks first using `find_block` before considering MCP.
+The platform has 480+ built-in blocks covering Google Workspace, social media,
+databases, and more. Only fall back to `run_mcp_tool` when no matching blocks
+exist and the service is in the known MCP servers list.
+
 ### Sharing files with the user
 After saving a file to the persistent workspace with `write_workspace_file`,
 share it with the user by embedding the `download_url` from the response in

autogpt_platform/backend/backend/copilot/sdk/mcp_tool_guide.md

@@ -20,7 +20,24 @@ Use these URLs directly without asking the user:
 | Cloudflare | `https://mcp.cloudflare.com/mcp` |
 | Atlassian / Jira | `https://mcp.atlassian.com/mcp` |
 
-For other services, search the MCP registry at https://registry.modelcontextprotocol.io/.
+For other services, search the MCP registry API:
+```
+GET https://registry.modelcontextprotocol.io/v0/servers?q=<search_term>
+```
+Each result includes a `remotes` array with the exact server URL to use.
+
+### Important: Check blocks first
+
+Before using `run_mcp_tool`, always check if the platform already has blocks for the service
+using `find_block`. The platform has hundreds of built-in blocks (Google Sheets, Google Docs,
+Google Calendar, Gmail, etc.) that work without MCP setup.
+
+Only use `run_mcp_tool` when:
+- The service is in the known hosted MCP servers list above, OR
+- You searched `find_block` first and found no matching blocks
+
+**Never guess or construct MCP server URLs.** Only use URLs from the known servers list above
+or from the `remotes[].url` field in MCP registry search results.
 
 ### Authentication
 

autogpt_platform/backend/backend/copilot/tools/run_mcp_tool.py

@@ -33,6 +33,9 @@
 # HTTP status codes that indicate authentication is required
 _AUTH_STATUS_CODES = {401, 403}
 
+# Status codes where the URL almost certainly doesn't host an MCP server
+_NOT_MCP_STATUS_CODES = {404, 405, 406}
+
 
 def _service_name(host: str) -> str:
     """Strip the 'mcp.' prefix from an MCP hostname: 'mcp.sentry.dev' → 'sentry.dev'"""
@@ -185,9 +188,22 @@ async def _execute(
                 # Server requires auth and user has no stored credentials
                 return self._build_setup_requirements(server_url, session_id)
             logger.warning("MCP HTTP error for %s: %s", server_host(server_url), e)
+
+            host = server_host(server_url)
+            if e.status_code in _NOT_MCP_STATUS_CODES:
+                error_msg = (
+                    f"No MCP server found at {host} (HTTP {e.status_code}). "
+                    "This URL does not appear to host an MCP server."
+                )
+            else:
+                error_msg = f"MCP server at {host} returned HTTP {e.status_code}."
             return ErrorResponse(
-                message=f"MCP server returned HTTP {e.status_code}: {e}",
+                message=error_msg,
                 session_id=session_id,
+                # Raw HTTP detail goes in `error` so the frontend can
+                # render it in a collapsible/de-emphasised block instead
+                # of dumping the full body (which may be an HTML page).
+                error=f"HTTP {e.status_code}: {str(e)[:300]}",
             )
 
         except MCPClientError as e: