Expose call call/transcription and recording endpoints in the HighLevel MCP server (GET /conversations/locations/:locationId/messages/:messageId/transcription)
L
Luke Brannon
Call transcripts are visible in the HighLevel web UI, but they are not exposed by the official HighLevel MCP server. The REST API already has dedicated endpoints:
• GET /conversations/locations/:locationId/messages/:messageId/transcription
• Get Recording by Message ID
• Download transcription by Message ID
The official MCP server ( https://services.leadconnectorhq.com/mcp/ ) currently includes 36 tools and covers conversations, but it does not provide any tool that wraps these transcription or recording endpoints. As a result, AI agents connected via MCP cannot retrieve call transcript text or recordings, even though the data exists in the account.
I need MCP tools that:
• Return the full transcript text for a given message ID
• Return the recording URL/file for a given message ID
• Optionally download the transcript file
This is a high-impact feature for Voice AI, call automation, compliance, and knowledge-base workflows. Please add MCP tools for:
• get_message_transcription(messageId, locationId)
• get_message_recording(messageId, locationId)
• download_transcription(messageId, locationId)
Log In
L
Luke Brannon
Thanks for the heads up Jay Woodcook!
J
Jay Woodcook
Worth flagging that this shipped. HighLevel Support On the current MCP (multi sub-account OAuth
connection) all three operations now exist: get-message-transcription,
download-message-transcription, and get-message-recording.
I tested get-message-transcription today against a live sub-account and it is byte-identical to the REST endpoint. I ran the returned segments through the same
renderer we use on the REST response and got the same 1404 characters, same speaker
indices, same sentence splits. Speaker separation is intact. It works.
get-message-recording is the one still broken. It returns 200 with real WAV audio,
but the registry declares the endpoint as application/json, so the bytes get UTF-8
decoded on the way out and 52 percent of the file comes back as U+FFFD. The RIFF
size field is destroyed, so the file is not recoverable. Filed separately here:
Luke Brannon Matthew Miller Rand Alexander
So looks like two of your three asks are done, and the third (my new bug post) needs a content type fix rather than a new tool.
M
Matthew Miller
This would directly support agent-ready follow-up workflows. AI needs access to call transcripts and recordings so it can prepare accurate follow-up emails and CRM notes. It should work alongside exact contact search, conversation retrieval, and unsent draft creation.