What is this?
MCP has two layers of errors. Connection errors are plain HTTP responses from the server before any tool runs (a missing key, no channel, a plan that doesn’t include MCP). Tool problems are reported back to the assistant as the result of the tool call, so it can explain them in plain words.
What can I do here?
- Fix a connection that won’t start
- Recognize why a tool is refused or does nothing
- Know what the assistant is telling you
How to use it
Connection errors
These use the standard envelope { "success": false, "error": "...", "message": "..." }.
When a tool can’t do what was asked, its result carries a plain message the assistant reads out. Common ones:
- Not found — an id the assistant used doesn’t exist in this channel (for example AI agent not found in this channel) or belongs to another platform. Ask it to list first.
- Nothing to change — an update was sent with no fields.
- Refused by a rule — for example a closed ticket can’t be replied to, or an automation that is on must be paused before its flow is replaced.
- Needs your approval — Write tools run only after your MCP client’s approval step. If nothing happens, check for a pending prompt.
- Not offered — a tool you lack permission for never appears in the list at all.
Troubleshooting / Technical Notes
- A write never seems to happen. Your client (Claude Code, Claude Desktop, Cursor, VS Code) asks you to approve each Write tool call — look for a pending approval prompt.
- The assistant says it can’t delete or cancel. By design: destructive actions are never exposed over MCP and stay in the dashboard.
403 suddenly after it worked. The plan or the key owner’s permissions changed — permissions are re-checked on every call.
Last modified on October 29, 2026
