Core Concept & Value Proposition

MeetingMCP transforms your organization's entire meeting history into a live, agent-queryable knowledge layer exposed as an MCP (Model Context Protocol) server. Instead of transcripts sitting siloed in Fireflies or Otter, your meeting corpus becomes something any AI agent, LLM client (Claude Desktop, Cursor, custom agents), or internal chatbot can query in real time. Ask "What did we decide about the enterprise pricing model last quarter?" and get a sourced, grounded answer with a link to the exact meeting clip.

The core insight: meeting transcripts are the most information-dense, chronologically rich, and universally ignored knowledge asset inside every company. MeetingMCP makes them alive.

Problem Being Solved

• Meeting transcripts are siloed and unsearchable at a semantic level across tools
• Teams spend hours manually prepping for recurring meetings by digging through old notes
• Action items and decisions made in meetings disappear into the void
• No existing tool exposes meeting history as a first-class data source for AI agents
• Enterprise meeting intelligence tools (Gong, Chorus) are priced out of reach for small teams

MVP Feature Set

• Transcript Ingestion: Connect one source — Fireflies.ai API, Otter.ai export, or direct VTT/SRT/text file upload
• Embedding Pipeline: Auto-chunk and embed transcripts using OpenAI or a local model, stored in a lightweight vector DB (pgvector on Supabase)
• MCP Server Endpoint: Expose a standards-compliant MCP server that any Claude Desktop or agent framework can point to — returns semantically relevant transcript chunks with meeting metadata (date, attendees, topic)
• Faceted Web Search UI: Simple web interface for direct search by date range, attendee, keyword, or topic — instant value even without an AI agent
• Slack Bot (stretch): /ask-meetings what did we decide about X? command that queries the MCP layer and returns a cited answer

Tech Stack (Small Team Friendly)

• Next.js + Vercel for frontend and API routes
• Supabase (pgvector) for storage and vector search
• OpenAI embeddings + GPT-4o for synthesis
• MCP SDK (Anthropic open-source) for the server layer
• Resend for onboarding/alert emails

Unique Selling Points

• First MCP-native meeting intelligence tool — rides the agent infrastructure wave directly rather than building another chatbot UI
• Bring your own transcripts — no new recording infrastructure needed; works with tools teams already use
• Privacy-respectable — can be self-hosted or run fully on Vercel with data never leaving the customer's Supabase instance
• Tiny surface area, massive leverage — a 2-person team can ship the core in 3–4 weeks using modern AI coding tools

Business Model

• Freemium SaaS: Free tier for up to 50 meetings/month, paid plans at $29/month (teams up to 10) and $99/month (unlimited + multiple integrations)
• Usage-based upside: Charge for API call volume above a threshold for teams running heavy agent workflows
• Target customer: Seed-to-Series B startups, product/engineering teams, and solo consultants who live in meetings and need their history to work for them

Go-To-Market

• Launch on Product Hunt with a live demo connecting to a public mock transcript corpus
• Post in AI Engineer / smol.ai community — directly aligned audience already using MCP tooling
• SEO content targeting "search meeting transcripts AI" and "MCP server for meetings"
• Partner/integration listing on Fireflies.ai and Otter.ai marketplaces

Potential Challenges & Mitigations

• MCP adoption still early: Mitigated by also offering a standard REST/chat API so the product works even without an MCP client — MCP is the differentiator, not the dependency
• Transcript data sensitivity: Offer one-click Supabase self-hosting and clear data deletion; make privacy a feature, not an afterthought
• Crowded transcript space: Positioning is explicitly not as a transcription tool but as an agent data layer — a fundamentally different category that Fireflies/Otter can't easily replicate without breaking their own positioning

Core Concept & Value Proposition

CasePilot is a lightweight AI copilot built specifically for social assistance caseworkers. It joins case meetings (phone, in-person via mobile mic, or video), transcribes the conversation in real time, automatically drafts the required case note or intake form, flags missing documentation, and proactively surfaces the client's next renewal or compliance deadline — all before the caseworker closes the session. The core promise: a caseworker finishes a 30-minute client meeting and the paperwork is already 80% done.

Social assistance agencies are chronically underfunded and understaffed. Caseworkers spend 40-60% of their time on documentation instead of direct client service. CasePilot attacks that ratio directly, giving each caseworker back 2-3 hours per day without requiring any technology change on the client's side.

The MVP (What Gets Built First)

The MVP is ruthlessly scoped to three features:

• Meeting Transcription: A web app (mobile-friendly) with a single "Start Session" button. The caseworker taps it at the start of a client call or meeting. It records and transcribes audio using Whisper/AssemblyAI, identifying speaker turns.
• Auto-Draft Case Note: After the session ends, an LLM (GPT-4o or Claude) processes the transcript against a configurable case note template (the agency pastes in their standard form fields during onboarding). It outputs a pre-filled draft the caseworker reviews, edits, and approves in under 2 minutes.
• Deadline Alert: The system parses the transcript for any benefit names, program enrollments, or dates mentioned, cross-references a simple rules table (e.g., SNAP renews every 12 months), and displays a "Next Action" card with the client's upcoming deadline and a one-click reminder that fires an SMS or email to both the caseworker and client.

The entire MVP runs on: Next.js + Vercel (frontend/backend), AssemblyAI or Whisper API (transcription), OpenAI API (form drafting), Resend (email reminders), Twilio (SMS reminders), and a simple Postgres database on Supabase. Two engineers can ship this in 6-8 weeks.

How AI Is Used

• Generative AI (LLM): Converts raw meeting transcripts into structured, agency-formatted case notes. Adapts to each agency's specific form templates via a simple system prompt configuration — no fine-tuning required.
• Speech-to-Text AI: Real-time or near-real-time transcription of case meetings, with speaker diarization to distinguish caseworker from client.
• Proactive Rule-Based + LLM Hybrid: Deadline detection combines hard-coded program renewal rules with LLM extraction of program names and dates mentioned in conversation — making it robust without requiring a complex knowledge graph on day one.

Unique Selling Points

• Zero client-side friction: Clients need no app, no smartphone, no tech literacy. The AI only touches the caseworker's workflow.
• Template-agnostic from day one: Agencies onboard by pasting their existing form templates. No custom integration required for MVP.
• Privacy-first architecture: Audio is processed and immediately discarded; only the transcript and draft note are stored, encrypted at rest. Agencies retain full data ownership — critical for government procurement.
• Sells to the budget holder, not the end user: Agency directors and county administrators are desperately seeking efficiency metrics. CasePilot can show "X hours of caseworker time saved per month" in a simple dashboard — exactly the ROI framing needed for government budget approval.
• Expansion path is clear: After MVP, add a local resource matching module (match client needs from transcript to a curated database of local nonprofits/shelters), multi-language support, and integrations with common case management systems (Apricot, Salesforce Nonprofit, Dynamics 365).

Go-To-Market

• Target small-to-mid county social services departments and nonprofit social assistance agencies (200-2,000 staff) — faster procurement cycles than federal agencies.
• Pricing: $30-50/seat/month. A 50-caseworker agency = $1,500-2,500 MRR per customer. 20 customers = $30-50K MRR.
• Land via direct outreach to county social services directors and nonprofit operations leads. Offer a free 30-day pilot with a single team of 5 caseworkers — the time-saved data sells the full contract.
• Leverage federal and state modernization grants (Title IV-E, SSBG, state digital equity funds) as a co-selling angle for budget-strapped agencies.

Potential Challenges & Mitigations

• Data privacy / HIPAA concerns: Mitigate by pursuing SOC 2 Type I early, offering a Business Associate Agreement, and keeping audio ephemeral. Many small agencies are not technically HIPAA-covered but appreciate the posture.
• Government procurement cycles: Mitigate by targeting nonprofit social assistance agencies first (faster decisions), then use those case studies to enter county/state contracts.
• Caseworker adoption resistance: Mitigate by making the UX feel like a simple voice memo app, not enterprise software. The caseworker's first session should produce a draft note better than what they'd write themselves — that's the hook.
• LLM hallucination in form fields: Mitigate by requiring caseworker review/approval before any note is saved, and highlighting LLM-generated fields visually so they are never silently accepted.