No install required. Recap hosts the MCP server for you. Add one URL and your API key to your AI client config. That's it.
Your AI agent is smart. It can write code, analyze data, draft emails. But ask it what your client said last Tuesday and it draws a blank. Every session starts from zero.
This is the core limitation of LLM-based agents: they have no persistent memory of real-world conversations. No access to what was actually said, agreed to, or promised in meetings, calls, and interviews.
Model Context Protocol (MCP) fixes this. In this guide, we'll show you how to connect Recap's hosted MCP server to Claude, ChatGPT, Cursor, or any MCP-compatible client — giving your AI access to a searchable archive of every voice conversation you've recorded.
What Is MCP? (30-Second Version)
Model Context Protocol is an open standard created by Anthropic that defines how AI tools connect to external data sources. Think of it as USB-C for AI agents — one protocol, any data source.
Before MCP, every AI integration was custom. You'd write a plugin for ChatGPT, a different extension for Claude, another adapter for your custom agent. MCP standardizes this into a single interface:
- Tools — functions the AI can call (search, fetch, create)
- Resources — data the AI can read (files, database records)
- Prompts — reusable prompt templates
An MCP server exposes these capabilities. An MCP client (Claude Desktop, Claude Code, Cursor, your custom agent) connects to the server and makes them available to the LLM.
The key insight: the MCP server is just a thin adapter. The real value is in the data behind it.
What a Conversation Memory MCP Server Does
Recap is a voice recording platform that transcribes, summarizes, and indexes every conversation you capture. The MCP server is a bridge between that archive and your AI agent.
Once connected, your agent can:
- Search across every recording you've ever made — by keyword, speaker, or topic
- Pull transcripts with speaker labels, timestamps, and AI-generated chapters
- Query contacts to see every conversation with a specific person over time
- Surface action items — commitments and follow-ups extracted by AI
- Read call quality scores if you're using Recap for team call review
Your agent goes from "I don't know what happened in that meeting" to "Sarah mentioned the budget was approved at 20:45 in your March 15th call — here's the exact quote."
Quick Start: Connect in 2 Minutes
Recap hosts the MCP server for you. No packages to install, no code to clone. Just a URL and an API key.
Step 1: Get your API key. Go to Settings > Developer in the Recap app and generate one. Keys start with recap_sk_ and are shown once — save it somewhere safe.
Step 2: Add this to your AI client config:
{
"mcpServers": {
"recap": {
"type": "http",
"url": "https://recap.cx/mcp",
"headers": {
"Authorization": "Bearer recap_sk_your_api_key_here"
}
}
}
}For Claude Desktop, paste this into ~/Library/Application Support/Claude/claude_desktop_config.json on macOS. For Claude Code, add it to your project's .mcp.json. For Cursor, add it in Settings > MCP.
Step 3: Restart your client. You'll see the Recap tools appear automatically. Start asking questions about your conversations.
Available MCP Tools
The server exposes 7 tools. Here's what each one does, with example calls and responses.
recap.search — Full-Text Search Across All Recordings
Find any moment from any conversation by keyword.
Tool call: recap.search({ query: "budget approval" }){
"results": [
{
"id": "AbC123xYz",
"title": "Q2 Planning with Sarah",
"date": "2026-03-15",
"duration": "34:12",
"matchingExcerpt": "...Sarah confirmed the budget was approved by finance last Friday...",
"matchTimestamp": 1245.3,
"contact": "Sarah Chen"
}
],
"total": 3,
"query": "budget approval"
}The matchTimestamp is in seconds — you can link directly to that moment in the recording.
recap.recent — Last N Recordings
Get a quick overview of recent conversations.
Tool call: recap.recent({ days: 7 })Returns the last 7 days of recordings with titles, dates, durations, and summaries. Useful for daily briefings or "what happened this week" queries.
recap.recording — Full Recording Detail
Pull the complete transcript, summary, chapters, and action items for a single recording.
Tool call: recap.recording({ id: "AbC123xYz" }){
"id": "AbC123xYz",
"title": "Q2 Planning with Sarah",
"date": "2026-03-15",
"duration": "34:12",
"speakers": ["You", "Sarah Chen"],
"summary": "Discussed Q2 budget allocation. Sarah confirmed finance approved the $50K expansion budget. Agreed to schedule vendor demos next week.",
"actionItems": [
"Follow up on budget proposal by Friday",
"Schedule vendor demo with Acme Corp"
],
"chapters": [
{ "title": "Budget Discussion", "start": "20:45", "summary": "Finance approved $50K expansion budget" },
{ "title": "Vendor Evaluation", "start": "28:10", "summary": "Comparing Acme vs. Globex proposals" }
],
"transcript": "Full transcript with speaker labels and timestamps..."
}recap.contacts — All Contacts
List everyone you've recorded conversations with.
Tool call: recap.contacts()Returns contact names, total recording count, and last conversation date. Your agent can use this to answer "who have I talked to recently?" or "how many calls have I had with this person?"
recap.contact_timeline — Conversation History with One Person
See every conversation with a specific contact, ordered chronologically.
Tool call: recap.contact_timeline({ contactId: "sarah-chen-id" })Returns a timeline of all recordings involving that contact — dates, titles, summaries. This is what makes the "brief me on everything with this client" use case work.
recap.sales.action_items — Open Commitments
Get action items extracted from recent recordings.
Tool call: recap.sales.action_items()Returns a flat list of commitments, follow-ups, and to-dos that were mentioned in conversations, with the source recording and date. Your agent can cross-reference these against your calendar or task list.
recap.sales.score — Call Quality Scorecard
If you're using Recap's call quality scoring (rubric-based evaluation with AI), pull the scorecard for any recording.
Tool call: recap.sales.score({ id: "AbC123xYz" })Returns scores per criterion, strengths, areas for improvement, and an overall grade. Useful for building QA automation on top of call recordings.
Use Cases
Pre-Meeting Prep Agent
Build an agent that runs before every calendar event:
"I have a meeting with Sarah Chen in 30 minutes.
Brief me on our last 3 conversations, any open action items,
and what I promised to follow up on."The agent calls recap.contact_timeline, then recap.sales.action_items, and synthesizes a briefing. Takes about 4 seconds. Replaces 20 minutes of scrolling through notes.
Customer Support Agent
Connect Recap's MCP server alongside your CRM's MCP server:
"Customer John Park is on the line. He called last week about
a billing issue. What did we tell him?"The agent searches for the prior conversation, pulls the transcript, and gives your support rep the exact context — no "can you remind me what this is about?"
Personal Knowledge Agent
Use Recap as a voice-first note-taking system:
"What did I decide about the API architecture?
I was talking through it with Mike on Wednesday."Your agent searches your recordings, finds the discussion, and returns the decision with context. Your conversations become a searchable knowledge base.
Automated Call Review Pipeline
For teams doing call quality monitoring:
# Pseudo-code: score all calls from this week, flag failures
recent = recap.recent(days=7)
for call in recent:
score = recap.sales.score(call.id)
if score.overall < 70:
notify_manager(call, score)Building on Top of Recap's MCP Server
The MCP server is hosted by Recap — you don't need to run anything locally. It's a thin adapter over Recap's REST API that exposes tools via the Model Context Protocol.
If you want to build a custom agent that uses conversation memory, here's the pattern:
import Anthropic from "@anthropic-ai/sdk";
const client = new Anthropic();
// Your agent's system prompt references Recap tools
const systemPrompt = \`You are a meeting prep assistant.
Before any meeting, use the recap.contact_timeline and
recap.sales.action_items tools to brief the user on context
and open commitments with that contact.\`;
// Claude automatically discovers and calls MCP tools
// when they're configured in the client
const response = await client.messages.create({
model: "claude-sonnet-4-20250514",
max_tokens: 1024,
system: systemPrompt,
messages: [
{
role: "user",
content: "I'm meeting with Sarah Chen in 30 minutes. Brief me."
}
]
});The MCP protocol handles tool discovery, schema validation, and execution. You write the prompt. Claude figures out which tools to call and in what order.
Why the Data Is the Moat
Let's be direct about this: an MCP server is not hard to build. The protocol is well-documented, the SDK is solid, and wrapping a REST API in MCP tools is a weekend project.
The hard part is what's behind the server.
Recap's value is the indexed conversation archive — months or years of transcribed, speaker-labeled, chapter-segmented, AI-summarized recordings. Every new conversation makes the system more useful. Your agent gets smarter not because the code improves, but because the data grows.
This is the flywheel for any MCP server: the protocol is the plumbing, the data is the product. If you're building MCP tools, think about what data you're making accessible, not just what endpoints you're wrapping.
Pricing and Free Tier
Recap offers a free tier so you can test the MCP server integration without commitment:
- Free: 30 min/month, all features including transcription, summaries, and search
- Pro ($29/mo): Unlimited recordings (1,500 min/mo), MCP server access, priority transcription
- Expert ($49/mo): Pro + call scoring + advanced AI templates
The MCP server is available on all paid plans (Pro and Expert). The API key is generated from your Recap account — no separate developer account needed.
FAQ
Do I need to install anything?
No. Recap hosts the MCP server for you. You just add a URL and your API key to your AI client's config file. No packages, no terminal commands, no dependencies.
How does authentication work?
API keys are generated in Settings > Developer in the Recap app. Each key is scoped to your organization. Keys start with recap_sk_ and are passed as a header in your MCP config. The server validates the key on every tool call.
What are the rate limits?
60 requests per minute per API key. This is generous for agent use — even an aggressive pre-meeting prep flow makes maybe 5-10 calls. If you're building batch processing, add a small delay between calls.
Where is my data stored?
Recordings and transcripts are stored in Recap's infrastructure (AWS S3 for audio, MongoDB for metadata and transcripts). All API calls go over HTTPS. Your API key is only sent to Recap's servers — never to a third party.
Does it work with tools other than Claude?
Yes. Any MCP-compatible client works — Claude Desktop, Claude Code, Cursor, Windsurf, ChatGPT, and any custom agent built with the MCP SDK. The protocol is client-agnostic.
What formats are supported for recordings?
Recap accepts audio (MP3, WAV, M4A, OGG, FLAC, WebM) and video (MP4, WebM, MOV) up to 500MB. Video files have their audio extracted automatically. The MCP server returns transcripts and metadata — not the raw audio files.
Can I write data back through the MCP server?
The current MCP server is read-only. You can search, fetch, and analyze — but not create recordings or modify transcripts. Write capabilities (bookmarks, comments, speaker corrections) are on the roadmap.
Ready to give your AI a memory?
Get your API key and connect in under 2 minutes. Your AI will have access to every conversation you've ever recorded.