C

C — Moderate risk signals Graded October 16, 2026Several concerns worth a closer look before you install.Based on public adoption and transparency signals as of October 16, 2026. Our opinion, not a code audit or security guarantee.

App connector · Web & browsing, Cloud & infrastructure

vexa

Open-source meeting transcription API for Google Meet, Microsoft Teams & Zoom. Auto-join bots, real-time WebSocket transcripts, MCP server for AI agents. Self-host or use hosted SaaS.

Bottom line: Signal-based grade (3/7 points): well starred on GitHub, public source code. This reflects adoption and transparency, not a code review.

Popularity score
39/100
Rank
#351 of 500
GitHub stars
2.9k

What it does

Vexa is an open-source meeting transcription API that automatically joins Google Meet, Microsoft Teams, and Zoom calls to generate real-time transcripts via WebSocket. It provides an MCP server interface for AI agents and can be self-hosted or run as a managed SaaS service.

Who it’s for

Builders integrating meeting intelligence into AI workflows, and enterprises needing automated transcription with deployment flexibility. Executives evaluating whether to adopt transcription automation for compliance, knowledge capture, or agent-driven workflows.

Copy-paste setup

Check these steps against the project’s own instructions before running anything.

Verify against the repo README at https://github.com/Vexa-ai/vexa (opens external site) for the latest instructions. Likely approaches include Docker self-hosting or connecting to the hosted SaaS via API credentials. For Claude Desktop integration as an MCP server, check the repo for an mcpServers configuration example.

Risks

  • Authentication scope: Bot credentials for Teams/Meet/Zoom require broad meeting access; compromised tokens could allow unauthorized call joining and recording.
  • Data exfiltration: Transcripts flow through the service (self-hosted or SaaS); ensure retention policies and encryption-in-transit align with data governance requirements.
  • Prompt injection: Transcribed user speech could contain adversarial input; downstream AI agents should validate and sanitize transcript content.
  • Supply chain: Relies on third-party meeting APIs (Google, Microsoft, Zoom); service disruptions or API changes affect availability.
  • Compliance gaps: Ensure usage complies with meeting platform ToS and local recording/consent laws; SaaS variant involves third-party data processing.

How to stay secure

Least-privilege permissions

  • Limit bot credentials to read-only transcript access — Configure meeting platform tokens (Teams, Zoom, Google Meet) to transcribe only, blocking the bot from starting/ending calls or changing settings
  • Isolate credentials by meeting type — Create separate bot accounts for Teams, Zoom, and Google Meet; if one credential leaks, attackers can only access that single platform
  • Rotate credentials every 90 days — Regularly replace bot tokens before they're compromised; store old credentials in a locked archive for audit purposes
  • Restrict AI agent access to necessary transcripts only — Don't grant agents blanket access to all meetings; only share transcripts they need to process

Network isolation

  • Self-host the service inside your corporate network — Avoid the SaaS version if handling sensitive meetings; this keeps transcript data from leaving your infrastructure
  • Encrypt transcripts in transit — Ensure all data moving to/from the transcription service uses TLS/HTTPS encryption (verify this is enabled in settings)
  • Control outbound connections — If self-hosting, use a firewall to block unexpected traffic to external APIs; only allow intentional calls to Google, Microsoft, and Zoom services
  • Separate AI agent networks — Don't let downstream AI systems access the raw transcription service directly; route all requests through an approval layer first

Monitoring

  • Log all transcript access — Record who requested which meetings' transcripts and when; review logs weekly for unusual patterns
  • Alert on credential usage anomalies — Set up notifications if bot tokens are used outside normal meeting hours or from unexpected locations
  • Track transcript retention — Confirm old transcripts are deleted per your data policy (e.g., "delete after 90 days"); audit quarterly
  • Monitor API call volumes — Watch for sudden spikes in transcript requests, which could indicate a compromised credential being abused

Red flags to watch for

  • Unexpected meetings in transcript logs — If the bot joins calls you didn't schedule, a credential has likely leaked; revoke it immediately
  • Transcripts appearing in unintended places — If meeting content shows up in chat channels, databases, or external systems you don't control, investigate data flow
  • Meeting platform access warnings — If Zoom, Teams, or Google flags unusual bot behavior (multiple simultaneous calls, rapid joining), pause the service and verify your setup
  • Delayed or missing transcripts — If transcription suddenly stops working, confirm the bot credential wasn't revoked by the meeting platform (often a sign of ToS violation)

Alternatives

  • Recall.ai – AI-powered meeting assistant with automatic transcription; managed SaaS model with stricter data handling.
  • Fireflies.io – Cloud-native meeting transcription and AI summaries; less flexible for self-hosting but strong compliance certifications.
  • Otter.ai – Consumer and enterprise transcription; mature, but primarily SaaS with limited MCP/agent integrations.
  • Symbl.ai – Conversation intelligence API; developer-friendly but acquired by Cisco and shifting product focus.

Sources