Troubleshooting

Common MCP server setup issues and how to fix them.

Common issues

The server isn’t appearing in my client

Check that your client supports HTTP transport. Not all MCP clients support Streamable HTTP — some only support stdio (local process) connections. The Trustgrid MCP server is HTTP-only. Verify your client’s MCP documentation.

Check the URL format. The URL must include the path prefix:

https://mcp.trustgrid.io/mcp          ✓ correct
https://mcp.trustgrid.io              ✗ missing /mcp
https://mcp.trustgrid.io/mcp/         ✓ trailing slash is fine

Check your domain. If your organization has a tenant-specific endpoint (e.g., mcp.acme.trustgrid.io), use that rather than the default mcp.trustgrid.io. If you use the portal at your-company-name.trustgrid.io, your MCP endpoint is mcp.your-company-name.trustgrid.io.


I’m getting 401 Unauthorized

Missing or malformed header. Confirm the Authorization header is present and uses the correct format:

# API token
Authorization: trustgrid-token YOUR_CLIENT_ID:YOUR_CLIENT_SECRET

A common mistake is using Bearer with an API token: Bearer clientId:secret won’t work. Use trustgrid-token clientId:secret.

OAuth loop. If the server returns a WWW-Authenticate challenge and your client doesn’t handle OAuth, it may retry indefinitely. Configure a static token in the headers instead.


I’m getting 403 Forbidden

Your credentials are valid but lack permission for the requested operation. Check:

  • The API key was generated for the correct user account
  • The user account has appropriate permissions in the Trustgrid portal
  • You’re using the correct endpoint path — /mcp/tools requires node service scopes that /mcp/read does not request

Tools work but return errors for specific nodes

Node is offline. Most diagnostic tools require the node to be reachable. If the node is disconnected from the control plane, test_* tools will time out or return an error.


Rate limit errors (429)

You’re hitting the per-token request limit for the active group. Limits are:

  • codemode: 10 requests / 60 seconds
  • read: 30 requests / 60 seconds
  • tools: 10 requests / 60 seconds

The response includes Retry-After header with the wait time in seconds. If you’re hitting limits consistently, consider using the read group instead of codemode for read-heavy workflows, or split high-frequency queries across multiple credentials.


OAuth won’t complete / browser auth fails

Check the MCP URL. OAuth flows redirect through portal.trustgrid.io or your tenant portal URL. If your network blocks the portal, the OAuth flow will fail. Use a static API token instead.

Check for cookie blockers. Some browser extensions block third-party cookies and can interfere with OAuth callbacks. Try in a private window.


The code tool says “not allowed” or isn’t available

The code tool is only in the codemode group (/mcp/codemode or /mcp). If you’re connected to /mcp/read or /mcp/tools, the code sandbox isn’t available. Change your client config to point at /mcp or /mcp/codemode.


Submit feedback

Found a bug, missing tool, or unexpected behavior? Open an issue at:

github.com/trustgrid/mcp-features

Helpful things to include in your report:

  • Client name and version — e.g., “Claude Code 1.2.3”, “Cursor 0.45”
  • Tool group / endpoint URL/mcp/codemode, /mcp/read, etc.
  • The tool that failed — the exact tool name (e.g., test_tcp_connectivity, code)
  • What you asked for — the prompt or tool call that triggered the issue
  • What you got back — the error message or unexpected response, including HTTP status codes if visible
  • What you expected — a brief description of the intended behavior
  • Auth method — OAuth or API token (don’t include the actual credential)

The more context you provide, the faster we can reproduce and fix it.