-
Notifications
You must be signed in to change notification settings - Fork 46
Troubleshooting
Apoorv Raj Saxena edited this page Jan 6, 2026
·
1 revision
- Confirm your server is wrapped:
- config uses
command: "mcp-reticle"andargs: ["run", ...]
- config uses
- Confirm the UI is running:
mcp-reticle ui
- Check terminal output for errors (server crash, missing binary, permission issues).
- Validate interception with log-only mode:
mcp-reticle run --log -- <server command...>
- If using proxy mode, confirm upstream + listen settings:
mcp-reticle proxy --help
- Port already in use: choose a different
--listenport - Upstream unreachable: verify the upstream is running and URL is correct
- Permissions: ensure the process can bind to the chosen port and read/write needed files
- Reticle captures stderr; inspect it in the UI.
- Run the server directly (without Reticle) to compare behavior.
- Ensure the same env vars and working directory as the client normally provides.
- Use token profiling to identify heavy methods/tools.
- Reduce tool schemas or response sizes where possible.
- Split toolsets across servers so you can isolate heavy ones.
Include:
- OS + architecture
- MCP client (Claude Desktop / Cursor / Cline / other)
- Transport (stdio / http+sse / streamable http / websocket)
- Server command (redact secrets)
- Minimal reproduction steps
- An export file (see Exports)
Reticle · MCP inspection & profiling · MIT licensed