Integration walkthrough · Updated 2026-05-22

Fireflies MCP + Claude: setup walkthrough and the 5 workflows that make your meeting corpus actually queryable

Fireflies publishes a beta MCP server documented at docs.fireflies.ai/getting-started/mcp-configuration with both an OAuth connector path (Claude/ChatGPT directories) and a direct API-key path. This walkthrough covers the setup, the Pro-tier requirement that gates API access, the 5 highest-leverage workflows, and the privacy considerations every operator should know before routing meeting transcripts through an LLM client.

Want to try Fireflies.ai?

Fireflies Pro ($18/user/mo annual) unlocks API + MCP access

OAuth via Claude/ChatGPT connector directories — one-click install. Or direct API key for Claude Code workspaces. Either way, your meeting corpus becomes queryable in chat.

Start with Fireflies →Affiliate link — StackSwap earns a commission if you sign up for Fireflies.ai. We only partner with tools we'd recommend anyway.

Step 1: Confirm your Fireflies plan supports API access

Fireflies gates API access (and therefore MCP) to paid tiers. The Free plan does meeting recording and AI summaries but doesn't include programmatic API. Pro at $18/user/mo billed annually is the entry point; Business at $29/user/mo expands integration limits; Enterprise adds advanced AI features. For most operators starting with Fireflies + Claude, Pro is the right tier.

Step 2: Pick the connection path (OAuth or direct API key)

Two options, pick based on which Claude surface you're using:

Step 3: Verify the connection

Smoke test in chat: "Using the Fireflies MCP, list my 10 most recent meetings." If the connector is healthy, Claude responds with your actual meeting list, with titles, dates, and participants. If the call fails: (a) verify the API key wasn't truncated on paste, (b) confirm the OAuth flow actually completed (the connector status in Claude should show "Connected"), (c) confirm your Fireflies plan actually includes API access. Fix in that order.

The 5 workflows that earn their install

1. Multi-meeting corpus queries (the flagship use case)

"Show me every mention of competitor X across the last 30 days of customer calls, with 2 minutes of surrounding context." The LLM walks your meeting corpus, finds the mentions, returns them grouped by call with timestamps. This is the workflow that previously required searching individual meetings one by one — collapsed into a single chat exchange.

2. Targeted single-meeting retrieval

"What did Acme's CTO say about implementation timeline in the May 12 call? Pull the exact quote and the 5 minutes of context around it." Speaker disambiguation plus timestamp-anchored retrieval, in chat. Useful for follow-up prep when you need the exact phrasing of what someone said.

3. Aggregate AI-summary analysis

"Walk last quarter's customer calls. What are the top 5 recurring objections? For each, pull a representative quote from a different meeting." The LLM queries Fireflies' AI summaries across the corpus, identifies patterns, returns themes with evidence. Hours of manual review compressed into a chat prompt.

4. Follow-up email drafts grounded in transcripts

"Draft a follow-up to Jane at Acme referencing the specific implementation concerns she raised in our May 12 call." The LLM has access to the actual transcript content, so the follow-up references real quotes and concerns — not the operator's reconstruction from memory. The quality improvement is meaningful when the operator runs 8-12 customer calls per week and can't remember the details of each.

5. Onboarding artifact generation for new hires

"Pull the 10 most-discussed objections from last quarter's calls. For each, identify which rep handled them best (judging by call outcome / next steps) and pull the direct quote of how they handled it." Generates a real-data-grounded onboarding artifact that previously required senior reps to handcraft from memory.

The privacy consideration — meeting content via LLM

Operator-level points worth marking:

The beta-status gotcha

Fireflies MCP is in beta. For interactive chat use this barely matters — the surface is stable enough for daily querying. For scheduled agent loops, the moving-target risk is real: tool surface may change, rate limits may shift, error semantics may evolve. Operator advice: build interactive workflows now, defer high-volume automation until the beta status moves to GA.

Where StackSwap MCP fits alongside Fireflies MCP

Fireflies MCP exposes your meeting corpus. StackSwap MCP exposes the cross-vendor GTM catalog — ~400 tools with monthly costs, AI-readiness scores, overlap pairs. Load both into the same Claude session: "summarize last week's customer calls" (Fireflies MCP) plus "what should our meeting-intelligence stack look like at our scale" (StackSwap MCP). One chat, two layers of insight.

Connect StackSwap MCP free →

Want to try Fireflies.ai?

Fireflies MCP + Claude makes your meeting corpus genuinely queryable

Pro tier ($18/user/mo annual), OAuth or API-key paths, 5-minute setup. Beta status is real; the leverage is real.

Start with Fireflies →Affiliate link — StackSwap earns a commission if you sign up for Fireflies.ai. We only partner with tools we'd recommend anyway.

FAQ

Two paths. (1) Connector directory — in Claude Desktop or claude.ai: Settings → Connectors → search for Fireflies → click install → complete OAuth in the browser. No key handling on your side. (2) Direct API key — generate a key in Fireflies (Settings → API → Developer Settings), then Settings → Connectors → Add custom MCP, paste the Fireflies MCP endpoint and API key. The OAuth path is the recommended default for chat surfaces; the direct-key path is the right choice for Claude Code workspaces or agent loops outside the official connector listings. Reference docs at docs.fireflies.ai/getting-started/mcp-configuration document both flows.

Yes — API/MCP access is gated to paid Fireflies tiers. The Free plan includes meeting recording and AI summaries but doesn't include programmatic API access, so the MCP isn't reachable there. Pro plan ($18/user/mo billed annually) is the entry point. Business ($29/user/mo) expands integration limits and Enterprise adds the advanced AI features. For most operators evaluating Fireflies + Claude, Pro tier is the right starting point — sufficient API access and the AI summary quality that makes the corpus useful.

Five workflows: (1) multi-meeting corpus queries — 'show every mention of competitor X across the last 30 days of customer calls with surrounding context'; (2) targeted single-meeting retrieval — 'what did Acme's CTO say about implementation timeline in the May 12 call'; (3) aggregate AI-summary analysis — 'top 5 recurring objections across last quarter's customer calls'; (4) follow-up email drafts grounded in real transcript content rather than the operator's memory; (5) onboarding artifact generation — 'pull the 10 most-discussed objections from last quarter and the handling that worked best.' All of this happens in chat against the actual recorded corpus, not summaries-of-summaries.

Fireflies MCP is in beta and rate limits are still calibrating. For interactive chat usage you'll rarely hit ceilings — the limits are designed for sustained programmatic access, not human-typing-speed conversations. For agent loops walking many meetings (e.g., 'analyze every call from the last quarter'), you can. Mitigation: throttle conservatively, batch queries into reasonable groups, and watch the chat for rate-limit errors during heavy fan-out. The beta status means error semantics may shift; the docs are the canonical source for current limits.

If you're using the OAuth connector path, the connector manages auth and you don't handle keys directly — the OAuth scope is what the authenticated user has access to in Fireflies. If you're using the direct API-key path, yes: create a scoped key for Claude use, named explicitly so the audit log is readable. Don't paste your admin key into a shared agent config. The LLM inherits whatever the key owner can see — scope = blast-radius control.

Transcript content routes through your LLM client when Claude queries it — that means the transcript text passes through Claude's infrastructure. Fireflies' own data-handling (SOC 2 Type II, GDPR) covers the platform side; review Anthropic's data-handling separately if compliance matters. For sensitive meetings (HR, legal, executive sessions), use Fireflies' meeting-level access controls to exclude them from API access. This is more robust than prompt-level guardrails — the MCP can't return content the API can't see. For most B2B sales motion content this is a non-issue; for regulated industries, validate the data-routing path before broad deployment.

Zapier routes through a workflow runner — Claude calls Zapier, Zapier calls Fireflies. Adds latency, a failure point, and a separate cost line. For chat-driven transcript querying, direct Fireflies MCP wins — single hop, no orchestration overhead. Zapier earns its place for event-driven workflows like 'when a Fireflies meeting summary is ready, post highlights to Slack and create a CRM activity.' Different problem, different tool. Most operator stacks use MCP for chat-driven querying and Zapier for event-driven downstream plumbing.

Yes to all three. claude.ai and Claude Desktop support the OAuth connector directory path — the cleanest install. Claude Code reads MCP config from your workspace .mcp.json or project config — the direct API-key path is more natural there. Same MCP underneath; the auth flow differs by surface. The transcript-side capability surface is identical across all three Claude clients — that's the design intent of the protocol.

Related reading

Canonical URL: https://stackswap.ai/fireflies-mcp-claude-integration. Disclosure: StackSwap is a Fireflies affiliate. Setup steps above are the same ones we use internally.