#My Agnet made a custom webhook tool call and it's result was : "Not authorised to use this tool"H

1 messages · Page 1 of 1 (latest)

nocturne wingBOT
#
AI-generated Answer for kishore_vb

Short answer: your agent is trying to call a tool it isn’t allowed to use. Fix it by ensuring the webhook tool is properly created, attached to the agent, and permitted under the current tools/approval model.

Checklist to resolve

  1. Confirm the webhook tool is added to the agent
  • In the Agent > Tools section, add/configure the Server (Webhook) tool. Agents can only call tools explicitly assigned to them. (elevenlabs.io)
  1. Use the new tool_ids mechanism (legacy prompt.tools is deprecated)
  • As of July 2025, tools must be referenced via prompt.tool_ids (and system tools via prompt.built_in_tools). Update any code using the old prompt.tools or the deprecated add-tool endpoint. (elevenlabs.io)
  1. If the tool is provided via an MCP server, check tool approval
  • In MCP settings, ensure the tool isn’t Disabled and is either auto‑approved or configured to request approval; otherwise the agent cannot execute it. (elevenlabs.io)
  1. Verify API key permissions
  • API keys are feature‑scoped by default. Make sure the key you’re using has access enabled for Conversational AI/Tools. (help.elevenlabs.io)
  1. Verify webhook authentication for the external API
  • If the external API requires auth, configure headers/secrets or OAuth in Workspace Auth Connections and link them to the tool. (elevenlabs.io)
  1. Inspect logs to confirm the exact failure
  • Check the Call History page for the precise error, or fetch the conversation details via the API. (help.elevenlabs.io)

Relevant docs for reference

If you’ve done all the above and still see “Not authorised to use this tool,” please contact Support with the agent ID, tool ID, and a recent conversation ID.

lofty nacelle
#

How to do 6) Inspect logs to confirm exact failure of my tool

nocturne wingBOT
#
AI-generated Answer for kishore_vb

Below are the officially documented ways to inspect what failed when your Agent tried to use a tool:

  • During the live session: listen for agent_tool_response WebSocket events. Each tool execution emits an agent_tool_response payload with tool_name, tool_call_id, tool_type, and is_error. Log these to confirm whether the tool invocation errored in real time. (elevenlabs.io)

  • After the call: fetch the conversation record via GET /v1/convai/conversations/:conversation_id and review the transcript and metadata. The transcript entries are your system-of-record for what the agent tried, when, and with what context. (elevenlabs.io)

  • Prefer automated post-call logging: enable Post-call Transcription webhooks for your workspace. ElevenLabs will POST a payload that includes the full conversation data (transcript, analysis, metadata). The example shows transcript turns that can include tool_calls and tool_results; use this to pinpoint the exact step and error. (elevenlabs.io)

  • If you don’t see logs: confirm that logging isn’t disabled. Enterprise “Zero Retention Mode” restricts logging for ElevenLabs Agents; if enabled, you won’t see request history/transcripts. (elevenlabs.io)

References:

  • Client events (agent_tool_response with is_error) — ElevenLabs Docs. (elevenlabs.io)
  • Get conversation details (transcript/metadata) — ElevenLabs API Reference. (elevenlabs.io)
  • Post-call webhooks (transcription payload with conversation data) — ElevenLabs Docs. (elevenlabs.io)
  • Zero Retention Mode (logging restrictions) — ElevenLabs Docs. (elevenlabs.io)

Sources:
Client events | ElevenLabs Documentation
Get conversation details | ElevenLabs Documentation
Post-call webhooks | ElevenLabs Documentation
Zero Retention Mode (Enterprise) | ElevenLabs Documentation
Client events | ElevenLabs Documentation
Get conversation details | ElevenLabs Documentation
Post-call webhooks | ElevenLabs Documentation
Zero Retention Mode (Enterprise) | ElevenLabs Documentation

lofty nacelle
#

expand more on
2) Use the new tool_ids mechanism (legacy prompt.tools is deprecated)