Best AI Meeting Note Takers with MCP and AI Agent Integration (2026)

When your meeting data lives in a silo, it stays a silo. You transcribe, summarize, extract insights—and then what? You copy the text into your AI assistant, manually prompt it to connect dots, and repeat. Model Context Protocol changes that equation.

MCP—the open standard introduced by Anthropic and now adopted across OpenAI, Google, and Microsoft—lets meeting tools expose their data to AI assistants through a standardized interface. Once connected, Claude can query your entire meeting history, retrieve action items, surface commitments, and draft follow-ups without you switching tabs or copying anything. Your meetings become searchable context inside the AI tools you already use.

By August 2026, the meeting tool space has stratified. The ones worth attention are not just the ones that transcribe—it's the ones that have built genuine MCP servers, not workarounds. Some are bidirectional (Claude can read meeting data and those tools can pull external context). Some are read-only (Claude reads your meetings, that's it). Some gate compliance features to paid tiers. The differences matter.

This guide covers 10 tools that have either native official MCP support, verified bidirectional MCP architecture, or production-ready third-party MCP implementations. We've tested setup complexity, verified compliance tier-gating, and compared what each tool actually exposes through MCP versus what marketing claims.

Quick Picks

For regulated teams that need governance and compliance data in every transcript: Fellow. It's the only tool pairing zero-day retention options, SOC 2 Type II, and HIPAA with an Anthropic-verified MCP connector and workspace-level administrator governance.

For sales and customer success: Avoma. Avoma offers deeper conversation analytics and was one of the first to publish a production-ready MCP server; it's now available through enterprise governance layers like Willow.How We Evaluated

MCP Implementation Depth

We distinguished between official native MCP servers (built and maintained by the vendor), verified third-party implementations (built by trusted partners and listed in official registries), and experimental or community implementations. Official implementations get priority in evaluation because they receive updates alongside core product releases and are supported in vendor documentation. Third-party implementations through partners like Willow or Gamut receive security reviews and enterprise governance layers. Community servers are noted but flagged as maintenance-dependent.

MCP Bidirectionality

Read-only MCP means Claude can query your meeting data—transcripts, summaries, action items—without modifying it. That covers most use cases: "What did we commit to last week?" "Summarize the Q3 planning call." Bidirectional MCP means the meeting tool can also pull context from external sources (email, CRM, documents) into its own search, and external tools can trigger actions within the meeting platform. Bidirectionality is powerful but adds complexity and deployment overhead. We noted which tools support each model and what that means for your workflow.

MCP Security and Data Retention

Some tools offer zero-day retention (ZDR) options that delete source recordings and transcripts immediately after processing, preserving only AI-generated summaries and action items. ZDR protections apply to core platform usage. Critically, ZDR does not always extend to MCP-connected data. If Claude has been granted access to your meeting library, ZDR constraints may not apply to what Claude can see or cache. We've noted which tools explicitly gate this behavior and which leave it ambiguous.

Cross-Platform Capture

Meeting tools differ in which platforms they support natively and which require workarounds. Some capture from Zoom, Teams, and Google Meet seamlessly. Others require browser extensions or bot participation. A few (Granola, Limitless) work at the system audio level and thus work across any platform, but that's a trade-off—you lose integration depth in exchange for flexibility. We've noted capture method for each tool because it affects whether MCP queries will surface all your meetings or a subset.

Pricing and Value at Scale

Most tools offer freemium models with tight usage caps (300 minutes/month, 25 meetings lifetime) and paid tiers starting $8–20/user/month. The pricing we've cited is as of August 2026; annual discounts typically reduce monthly costs 15–20%. For MCP specifically, some tools include MCP access on all plans; others gate it to Pro or Enterprise tiers. We've noted where MCP access is restricted because connecting an unpaid user to Claude can create unexpected bill shock.

Individual Tool Profiles

1. Fellow

MCP Implementation: MCP Server. Official, Anthropic-verified OAuth connector for Claude Desktop and Claude Code. Native to Claude's connector directory.

Bidirectionality: Yes. With write capabilities enabled (like creating meeting agendas), Fellow’s Model Context Protocol (MCP) server becomes bidirectional. Claude can query meeting notes, transcripts, AI summaries, action items, and calendar context through Fellow's MCP connector. Write access can be used to create meeting agendas from Claude. Fellow does not pull external context into Claude through this connector, though users can combine Fellow with other Claude connectors (like Google Drive or Linear) in the same conversation. Fellow also has separate, non-MCP integrations with 50+ tools including CRM, Jira, Linear, and HubSpot.

What MCP Exposes: Meeting summaries, transcripts, action items, custom tracker hits (compliance phrases, client commitments, regulatory triggers).

Capture Method: Botless (native Teams integration) or bot participant on Zoom/Google Meet. In-person and hybrid recording via speaker diarization.

Compliance: SOC 2 Type II, GDPR, HIPAA. Zero-day retention mode available; ZDR constraints apply to MCP-accessed data. BAA available for Enterprise. Domain-based recording blocks, customizable consent disclosures, information barrier policies for separation of duties.

Key Differentiator: Workspace-level administrator governance. Compliance teams define recording policies, retention schedules, and access permissions once; Fellow applies them across all teams and meeting types. Super Admin API exports audit-ready deletion logs. This is governance-first design, not an add-on.

Pros: Strong compliance controls; transparent tier-gating; Anthropic-verified connector means updates ship alongside Claude releases; zero-day retention protections extend to MCP data; Enterprise plans include HIPAA and BAA.

Cons: Smaller integration footprint than Otter or Fireflies on certain platforms (native Teams support is strength).

Pricing: Per-seat pricing; specific rates require contact. Free tier includes basic capture and 7-day retention. Pro and Business tiers start ~$10–20/user/month (annual billing) and include MCP access. Enterprise is custom and required for HIPAA.

Setup Complexity: 3 minutes. OAuth flow built into Claude Desktop and Claude Code connectors; no API keys or remote URLs to manage.

2. Otter.ai

MCP Implementation: Official bidirectional MCP server

Bidirectionality: Yes. Claude queries Otter meeting data; Otter AI Chat pulls context from Gmail, Google Drive, Notion, Jira, and Salesforce to enhance search. This bidirectional architecture makes meeting data portable and connectible to broader enterprise context.

What MCP Exposes: Full transcript search, summaries, speaker insights, action items. Otter's MCP implementation is one of the most feature-rich; Claude can filter by date, participant, keywords, and combine multiple queries.

Capture Method: Bot joins Zoom, Teams, Google Meet, and handles in-person via uploaded audio or phone dial-in. Otter auto-joins from calendar events (calendar integration optional).

Compliance: SOC 2 Type II (all plans). HIPAA and BAA available on Enterprise tier. GDPR compliant. No zero-day retention option; retention is configurable (30 days to indefinite) but source recordings are retained.

Key Differentiator: The breadth of what MCP exposes and bidirectional architecture. Otter AI Chat can pull meeting context alongside email and project data, making meeting history searchable within a unified query layer.

Important Context: Otter faced federal litigation filed in August 2025 alleging the tool recorded conversations without explicit consent in certain scenarios. The lawsuit is ongoing and consolidated. This is material context for regulated teams and those with consent-sensitive use cases. Otter's consent workflows have been updated since the filing, but the litigation remains active.

Pros: Feature-richest MCP implementation; bidirectional architecture; free tier includes unlimited transcription (with storage limits); strong multilingual support (works well with non-English audio).

Cons: Ongoing class-action litigation around consent creates trust concerns for some buyers; no zero-day retention option; audio quality in noisy environments can degrade transcription accuracy; free tier has generous storage limits but limited to 300 minutes/month; ZDR protections do not exist.

Pricing: Free tier: 300 minutes/month, storage limits. Pro: $8.33/user/month (annual) or $12.50/month (monthly), 1,200 minutes/month. Business: $20/user/month (annual), 6,000 minutes/month, unlimited file imports. Enterprise: Custom pricing, includes SSO, SCIM, admin controls, audit logging. MCP access available on all paid tiers.

Setup Complexity: 5–10 minutes. OAuth through Claude's connector directory or manual remote MCP URL configuration for enterprise deployments.

3. Fireflies.ai

MCP Implementation: Official MCP server.

Bidirectionality: Read-only. Claude queries Fireflies data; Fireflies does not pull external context into its own search through MCP (though Fireflies has separate non-MCP integrations with ATS, CRM, and communication platforms).

What MCP Exposes: Transcripts, summaries, speaker data, user information, and cross-meeting search. The Fireflies MCP exposes 8 core read-only tools; third-party partners like StackOne have extended implementations that include write operations (transcript privacy updates, soundbite creation).

Capture Method: Bot joins calls on Zoom, Teams, Google Meet. No botless mode; bot participation is required.

Compliance: SOC 2 Type II. Free tier includes unlimited transcription subject to storage and AI-credit limits; no HIPAA or BAA on free tier. Paid plans include GDPR compliance; HIPAA available on Enterprise tier.

Key Differentiator: Multilingual strength. Fireflies handles 30+ languages with high accuracy, making it the go-to choice for multilingual teams. Launched first in Claude's connector directory, giving it early adoption traction.

Pros: Strong multilingual support; unlimited transcription on free tier (unlike Otter's 300-minute cap); early MCP adoption and wide community use; cross-meeting search is powerful; integration breadth (50+ integrations including recruiting ATS); transparent on storage limits and AI credits.

Cons: Bot-required capture (no botless mode); ZDR protections do not extend to MCP-connected data, creating compliance risk for regulated teams; HIPAA only on Enterprise tier; customer support response times have drawn complaints in reviews; read-only MCP (no bidirectional architecture).

Pricing: Free plan: Unlimited transcription with storage and AI-credit limits. Pro: ~$10/user/month (annual) or $18/month (monthly). Business: ~$20/user/month. Enterprise: Custom. MCP access available on all tiers. Verify at fireflies.ai before purchasing; pricing updated frequently.

Setup Complexity: 3–5 minutes. OAuth connector in Claude Desktop; or manual remote MCP URL for enterprise setups.

4. tl;dv

MCP Implementation: Official native MCP server.

Bidirectionality: Read-only. Claude queries tl;dv meeting data. tl;dv does not pull external context through MCP (though integrations with Notion, Jira, Salesforce exist outside MCP).

What MCP Exposes: Meeting transcripts, summaries, participant data, and semantic search across full meeting history. tl;dv's implementation supports filtering by host, participant, date range, and keywords. One of the richest MCP implementations alongside Otter.

Capture Method: Native browser extension for Zoom, Google Meet, Teams; works across all three major platforms without requiring a bot join or system-level audio capture. Works with in-person/hybrid via manual upload.

Compliance: GDPR (EU data residency option available). SOC 2 Type II available for enterprise customers. No HIPAA or BAA. No zero-day retention option.

Key Differentiator: Platform breadth with seamless native capture. tl;dv's browser extension works identically across Zoom, Teams, and Google Meet, making it one of the few tools with truly universal platform support through a single mechanism. EU hosting option appeals to privacy-focused teams in regulated jurisdictions.

Pros: Truly cross-platform native capture (Zoom, Teams, Google Meet all identical UX); EU hosting option; semantic search quality is strong; free tier offers unlimited recordings with 3-month auto-delete (generous compared to 300-minute caps elsewhere); fast-moving team that shipped MCP early.

Cons: No HIPAA or BAA; no zero-day retention; read-only MCP (no bidirectional); free tier auto-deletes after 3 months, creating data-loss risk if you don't upgrade; smaller compliance footprint than Fellow or Otter.

Pricing: Free tier: Unlimited recordings, 3-month auto-delete, storage limits. Pro: ~$12/user/month, unlimited history retention. Business: ~$25/user/month. Enterprise: Custom. MCP included on all tiers. Annual billing typically offers 15–20% discount.

Setup Complexity: 5–10 minutes. Remote MCP URL configuration; OAuth for first connection, then cached.

5. Avoma

MCP Implementation: Official MCP server.

Bidirectionality: Read-only through base MCP. Willow's enterprise wrapper adds governance layers but does not modify bidirectionality (still read-only to external AI assistants).

What MCP Exposes: Meeting transcripts, AI-generated summaries, conversation analytics (filler words, monologue detection, competitive mentions), action items, and structured insights. Avoma's MCP is designed for sales and revenue operations teams; analytics output is detailed.

Capture Method: Bot joins Zoom, Teams, Google Meet. Handles in-person via audio upload.

Compliance: SOC 2 Type II. GDPR compliant. HIPAA and BAA available on Enterprise tier. When deployed through Willow (enterprise MCP gateway), requests run behind SSO (Okta, Azure AD), RBAC, and full audit trail.

Key Differentiator: Conversation analytics depth. Avoma doesn't just transcribe; it flags monologues, competitive threats, and engagement metrics. When queried through MCP, Claude can extract these insights alongside transcripts. Enterprise governance layer (Willow) adds SIEM integration (Splunk, Loki) for auditable MCP access.

Pros: Rich conversation analytics through MCP (not just transcripts); enterprise governance option through Willow eliminates security review friction; strong sales-team focus (CRM integrations, deal insights); SOC 2 Type II on all plans.

Cons: Bot-required capture; analytics depth can mean slower query performance in MCP calls; requires Willow for enterprise SIEM integration (Willow adds cost/infrastructure); HIPAA only on Enterprise tier; read-only MCP.

Pricing: Pro: ~$15/user/month. Business: ~$25/user/month. Enterprise: Custom. MCP included on Pro tier and above. Willow governance adds separate fees for enterprise control plane.

Setup Complexity: 5–10 minutes for base MCP; 10–20 minutes for Willow integration (SSO setup, RBAC configuration).

6. Granola

MCP Implementation: Community-driven third-party MCP server (not official Granola implementation). Granola itself does not publish or maintain an MCP server; third-party developers have created MCP wrappers around Granola's API.

Bidirectionality: Read-only (community implementation).

What MCP Exposes: Transcripts, summaries, and structured notes from Granola's hybrid human+AI note-taking system. The community implementation is less feature-rich than official servers but functional for basic querying.

Capture Method: Botless. Granola records audio locally on your device (Windows, Mac desktop app; not mobile). No visible bot joins calls. Records from system audio, works across any conferencing platform.

Compliance: No SOC 2, HIPAA, or BAA. Granola is designed for individuals and small teams, not enterprises with compliance requirements. Local-first architecture (recordings on your device, not cloud) appeals to privacy-focused users but provides minimal audit trail for regulated deployments.

Key Differentiator: True botless capture and hybrid note-taking (your handwritten notes merged with AI context). Local-first processing appeals to teams that distrust cloud recording.

Pros: Botless capture (no interruption to calls); local-first processing (your control); hybrid note-taking UX appeals to users who take their own notes; minimal infrastructure footprint; individual pricing ($18/month) is competitive.

Cons: MCP support is community-driven, not official—maintenance risk and no guaranteed updates; no compliance certifications; local-only recording limits team sharing and audit trails; smaller feature set than Otter, Fireflies, or tl;dv; free tier caps at 25 meetings lifetime (tight); no enterprise deployment path.

Pricing: Free tier: 25 meetings lifetime. Pro: $18/month individual. No team or enterprise plans. MCP access available through community implementation at no additional cost (but no SLA).

Setup Complexity: 10–15 minutes. Requires manual configuration of third-party MCP server; Granola doesn't provide official documentation, so setup relies on community tutorials and GitHub repos.

7. Read AI

MCP Implementation: Official MCP server in open beta (May 2026). Read AI MCP and REST API both available in beta; production readiness varies.

Bidirectionality: Read-only. Claude queries Read AI meeting data through MCP.

What MCP Exposes: Meeting summaries, transcripts, speaker engagement metrics, and action items. Read AI focuses on meeting engagement (speaking time, interruptions, sentiment); MCP exposes these analytics alongside transcript data.

Capture Method: Visible bot joins Zoom, Teams, Google Meet. Auto-joins based on calendar sync.

Compliance: SOC 2 Type II. GDPR compliant. HIPAA available on Enterprise tier. No BAA noted.

Key Differentiator: Meeting engagement analytics through MCP. Unlike tools focused purely on transcription, Read AI flags speaking patterns, engagement gaps, and meeting health metrics queryable through Claude.

Pros: Engagement analytics valuable for sales and management teams; MCP in open beta signals commitment to integration; free tier includes 5 meetings/month for evaluation; bot auto-joins from calendar (minimal manual setup).

Cons: MCP is still in open beta (not production-certified); bot visibility creates friction in sensitive calls; limited free tier (5 meetings/month vs. 300 minutes on Otter, unlimited on Fireflies); smaller integration ecosystem compared to Otter or Fireflies; read-only MCP; HIPAA only on Enterprise.

Pricing: Free tier: 5 meetings/month. Pro: ~$12/user/month. Business: ~$20/user/month. Enterprise: Custom, includes HIPAA. MCP included on all tiers but in beta; verify at read.ai.

Setup Complexity: 5–10 minutes. OAuth through Claude's connector directory; beta status means occasional API changes.

8. MeetGeek

MCP Implementation: Official MCP server availabl

Bidirectionality: Read-only.

What MCP Exposes: Meeting summaries, transcripts, participant insights, and action items. MeetGeek's MCP is functional but less feature-rich than Otter or tl;dv; query options are more limited.

Capture Method: Bot joins Zoom, Teams, Google Meet.

Compliance: SOC 2 Type II (verified). GDPR compliant. No HIPAA or BAA.

Key Differentiator: Workflow automation focus. MeetGeek is built for teams prioritizing complex workflow automation (routing action items to specific tools, conditional summaries, template-based outputs). MCP access is secondary to core product focus.

Pros: SOC 2 Type II on all plans; workflow automation is genuinely strong; MCP available on all tiers; affordable Pro pricing (~$12/user/month).

Cons: No HIPAA or BAA; read-only MCP; less feature-rich MCP implementation than Otter/tl;dv; smaller adoption base means community support is thinner; free tier is capped (limited to paid trial or contact sales).

Pricing: Free trial available. Pro: ~$12/user/month. Business: ~$20/user/month. Enterprise: Custom. MCP included on all tiers.

Setup Complexity: 5–10 minutes. Standard OAuth flow; straightforward setup.

9. Krisp

MCP Implementation: Official MCP server. Krisp's primary focus is audio quality enhancement; MCP is secondary to core noise-cancellation product.

Bidirectionality: Read-only.

What MCP Exposes: Meeting recordings, transcripts, and metadata. Krisp's MCP implementation is functional but focused on data retrieval rather than rich query options. Designed to expose what Krisp captures but not to provide deep search or analytics.

Capture Method: Desktop app records system audio (works across all platforms since it captures at system level, not integrating with specific conferencing tools). Also available as browser extension for Zoom/Teams/Google Meet.

Compliance: SOC 2 Type II. GDPR. No HIPAA or BAA.

Key Differentiator: Audio quality is the primary product. Krisp denoise engine is strong; MCP is bolt-on to expose those cleaned recordings to external AI tools. Best for teams prioritizing audio clarity in noisy environments.

Pros: Krisp's noise cancellation is genuinely good (independent tests show better performance than competitors' built-in noise suppression); system-level audio capture works with any platform; affordable (~$10/month); SOC 2 Type II.

Cons: No HIPAA or BAA; MCP is secondary focus (not as feature-rich as meeting tools whose primary focus is transcription); read-only MCP; noise cancellation alone is not a full meeting-recording solution (you still need a tool to handle transcription and summaries); free tier is limited.

Pricing: Free tier: Limited features, watermark. Pro: ~$10/month or $8.33/month (annual). MCP included on Pro tier.

Setup Complexity: 5–10 minutes. System audio configuration required (slightly more setup than cloud-native tools).

10. Limitless

MCP Implementation: Community-driven third-party MCP server. Limitless itself (the Pendant lifelog device) does not publish official MCP support; developers have created MCP wrappers around the Limitless API.

Bidirectionality: Read-only (community implementation).

What MCP Exposes: Lifelog entries (Pendant audio recordings), transcripts, and personal memory context. Limitless captures ambient audio; MCP exposes that data to Claude for querying across your personal and professional conversations.

Capture Method: Limitless Pendant (wearable device) records ambient audio continuously. Works everywhere; no platform-specific setup. Not designed specifically for meeting recording but captures audio in meetings as part of continuous lifelog capture.

Compliance: No SOC 2, HIPAA, or BAA. Limitless is designed for personal use; no enterprise deployment or audit trails.

Key Differentiator: Ambient capture outside meetings. Limitless records everything in your environment (conversations, background discussions, context). MCP lets Claude query this ambient context for personal productivity (not organizational governance). This is personal memory + AI, not meeting recording + AI.

Pros: Captures context outside formal meetings (hallway conversations, background discussions); seamless (no setup per call); privacy-first (device-level encryption); MCP lets Claude access your full lifelog context for personal workflows.

Cons: Not designed for team or regulated use; no compliance certifications or audit trails; community MCP (maintenance risk); ambient capture raises privacy concerns in some workplace environments; Pendant hardware required (~$99 + subscription); not a meeting recording tool in the traditional sense.

Pricing: Pendant hardware: ~$99 one-time. Subscription: ~$10–20/month depending on tier. MCP available through community implementation at no additional cost.

Setup Complexity: 20–30 minutes. Requires Pendant hardware setup, API key configuration, and third-party MCP server setup (GitHub repository). Higher bar than cloud-native tools.

Comparison at a Glance

MCP Architecture & Platform Support

Official Read-Only MCP (Anthropic-Verified or Directory-Listed):

• Fellow (Anthropic-verified OAuth connector)
• Fireflies (first in Claude directory)
• tl;dv (full platform support: Zoom, Teams, Google Meet)
• Avoma (official, available through Willow governance layer for enterprise)
• Read AI (open beta)
• MeetGeek (verified official server)

Official Read-Only MCP (Secondary Focus):

• Krisp (audio-focused; MCP is bolt-on)

Community/Third-Party MCP (Not Officially Maintained):

• Granola (community-driven wrapper around API)
• Limitless (community-driven wrapper around Pendant API)

MCP Extended Implementations:

• Fireflies + StackOne (adds write operations: privacy updates, soundbite creation)

Zero-Day Retention (ZDR) & Data Minimization

ZDR Available:

• Fellow (ZDR mode deletes source recordings/transcripts immediately; ZDR constraints apply to MCP-accessed data)

No ZDR Option:

• Otter (retention configurable but source recordings retained)
• Fireflies (retention configurable; ZDR does not extend to MCP-connected data)
• tl;dv (retention configurable; free tier auto-deletes after 3 months)
• Avoma (retention configurable)
• Read AI (no ZDR)
• MeetGeek (no ZDR)
• Krisp (retention configurable)
• Granola (local-first means your control; cloud retention not applicable)
• Limitless (ambient capture; no specific ZDR mode)

Pricing Entry Points & MCP Access

MCP Included on Free Tier:

• Otter (Free: 300 min/month + MCP access)
• Fireflies (Free: Unlimited transcription + MCP access; storage/AI-credit limits)
• tl;dv (Free: Unlimited recordings, 3-month auto-delete + MCP access)
• Avoma (Free: Contact sales; MCP on Pro tier+)
• Read AI (Free: 5 meetings/month + MCP in beta)
• MeetGeek (Free: Trial or contact sales)

MCP Requires Paid Tier:

• Fellow (Pro/Business required; HIPAA requires Enterprise)
• Krisp (Pro tier)
• Granola (Free tier has 25 meetings lifetime; Pro $18/month + community MCP)
• Limitless (Pendant + subscription required; community MCP)

Regulation & Litigation Context

Ongoing Legal Context:

• Otter: Federal class-action lawsuit filed August 2025 alleging non-consensual recording in certain scenarios; litigation ongoing. Consent workflows updated since filing; risk assessment required for consent-sensitive deployments.

Transparent Governance Design:

• Fellow: Workspace-level admin controls for recording policies, consent disclosure logging, and audit trails designed specifically for regulated teams.

How to Choose

You Need Compliance Documentation and Governance Controls: Fellow. It's the only tool shipping workspace-level admin governance, transparent tier-gating (HIPAA on Enterprise only, not implied company-wide), and ZDR protections that extend to MCP-accessed data. Compliance teams can define recording policies once; Fellow applies them everywhere.

You Need Meeting Engagement Analytics: Avoma or Read AI. Avoma has richer conversation analytics (filler words, competitive mentions, monologues) queryable through MCP; enterprise governance layer (Willow) adds SIEM integration. Read AI focuses on engagement metrics; MCP is newer (open beta) but maturing. Both appeal to sales/management teams tracking meeting health.

You Need Audio Quality First: Krisp. Best noise cancellation independent testing shows. MCP support is secondary but available. If your meetings happen in noisy environments (home offices, coffee shops, conference rooms with echo), Krisp's core product delivers value. MCP is functional but not feature-rich.

You Have Zero Budget or Need Unlimited Transcription: Fireflies (free tier, unlimited transcription, MCP included). Fathom also offers unlimited free tier if you don't need MCP specifically; Fathom has no official MCP but has strong free tier.

You're in a Regulated Industry and Need a Litigation-Free Vendor: Fellow. Otter's ongoing class-action lawsuit is material context. Fellow has no comparable litigation and positions compliance as first-class design, not an add-on. If consent/governance/audit trails are non-negotiable, Fellow is the only tool shipping workspace-level controls + ZDR + HIPAA (Enterprise) without legal baggage.

You're Evaluating Post-MCP Stateless Spec Release (July 2026): Expect updates to all tools' MCP implementations over the next quarter. The stateless MCP core simplifies deployment (no sticky sessions, shared state, or deep packet inspection). Cloud-native tools (Fellow, Otter, Fireflies, tl;dv) will adopt stateless spec releases within weeks. Community MCP servers (Granola, Limitless) may lag or require manual updates.

MCP FAQs

Is Fellow's MCP bidirectional?

Yes, with write capabilities enabled (like creating meeting agendas), Fellow’s Model Context Protocol (MCP) server becomes bidirectional.

Instead of operating strictly as a read-only integration, the MCP setup allows connected AI models to interact with your workspace in both directions:

• Read Capabilities (Inbound): The AI can search meetings, retrieve transcripts, fetch notes, pull action items, and view attendee lists.  
• Write Capabilities (Outbound): The AI can push changes back into Fellow, such as generating and creating meeting agendas, updating notes, or adding tasks.

Can I use MCP with Granola?

Yes, but support is community-driven. Granola does not publish official MCP. Third-party developers have built MCP wrappers around Granola's API. Setup is more complex (manual server configuration) and maintenance is not guaranteed. Use at your own risk for production deployments.

Which meeting tools work with ChatGPT through MCP?

Several: Fellow (Claude, ChatGPT, or Cursor can now tap directly into your meeting recaps, notes, and action items), Otter (bidirectional MCP works with ChatGPT), Fireflies (official MCP server available for ChatGPT), Read AI (official MCP available for ChatGPT), tl;dv (official MCP available for ChatGPT). Most meeting tool MCP servers are vendor-agnostic and work with any MCP-compatible client (Claude, ChatGPT, Cursor, VS Code, etc.). Check individual tool docs for ChatGPT-specific setup if official Claude connectors are your reference.

Does Otter's MCP include zero-day retention protection?

No. Otter does not offer zero-day retention. Retention is configurable (30 days to indefinite), but source recordings are always retained for the duration of your chosen policy. MCP access does not extend ZDR protections because Otter doesn't have ZDR as a feature. If your compliance requirement is immediate deletion of raw recordings post-processing, Fellow (with ZDR mode) is the alternative.

What data can I access through Fellow's MCP?

Meeting summaries, transcripts (redacted per your policies), action items, and custom tracker hits (compliance phrases, regulatory triggers, client commitments). Fellow's MCP does not expose raw audio files; all queries return structured, text-based data. Redaction rules applied at the platform level are preserved in MCP queries.

If I enable MCP access to my meeting data, does that bypass my retention policy?

Depends on the tool. Fellow explicitly applies ZDR constraints to MCP-accessed data (if you have ZDR enabled, MCP queries see only the protected subset). Fireflies, tl;dv, and others do not explicitly gate MCP to retention policies—once MCP access is granted, Claude can query data that would normally expire. Check your tool's documentation or contact support to confirm data-protection boundaries for MCP access.

Is MCP access audited?

Yes and no. Official MCP servers log connections (OAuth handshakes, authentication events). Some tools (Fellow, Avoma via Willow) log individual queries and include them in audit trails. Others (Fireflies, tl;dv) log that you've connected MCP but may not log individual queries. If query-level auditing is a compliance requirement, verify with your tool's support. Avoma through Willow includes full query-level audit trail + SIEM integration (Splunk, Loki).

Can I revoke MCP access if I change my mind?

Yes. Revoking MCP access in Claude's connector settings disconnects the server and Claude loses access to your meeting data. The process is reversible—reconnecting is the same OAuth flow. Some tools cache meeting data temporarily; check documentation for cache-clear timelines if you need immediate revocation.

Is there a performance difference between querying meetings through MCP vs. the native app?

Not significantly for most tools, but query-heavy workflows (automating analysis across hundreds of meetings) can hit rate limits on MCP servers. Fellow and Otter have higher rate limits for enterprise deployments. Fireflies and tl;dv have documented rate limits (typically 100–500 requests/hour for free/Pro tiers, higher for Enterprise). If you're building a workflow that queries thousands of meetings, budget for potential throttling or upgrade to Enterprise tier.

‍