Speaker labels via MCP connector — assigned names not coming through to Claude Code

General
1

Hi Pocket team,

I’m using the MCP connector to pull recordings into Claude Code and I’d like to understand the expected behaviour
for speaker labels.

In the Pocket UI I’ve assigned real names to the speakers in one of my recordings. When I fetch the same recording
through the MCP connector (using get_pocket_conversation), each segment in transcriptSegments still comes back
with the anonymous diarization label in the speaker field — SPEAKER_00, SPEAKER_01, etc. — instead of the names I
assigned.

A few questions so I can set my expectations correctly:

  1. Are user-assigned speaker labels meant to flow through the MCP connector, or is the MCP surface deliberately
    limited to the raw diarization output?
  2. If they are meant to flow through, is there a separate field I should be reading (e.g. on the recording object
    rather than on each segment), or a parameter I need to pass to get_pocket_conversation?
  3. Is there any propagation delay between assigning labels in the UI and them becoming visible through the
    API/MCP?

Happy to share a recording ID privately if that would help diagnose.

Thanks

2

hey @Cyllaris thanks for reporting this I’m checking this out!