MCP docs
MCP server
Connect Voicecape summaries to Claude, Cursor, and other MCP clients, so your assistant can search and read your team's meetings directly.
Same rp_ keys as the REST API — included on every plan, including Free. The ask tool needs a paid plan; the other seven run on Free.
Endpoint
The MCP server speaks Streamable HTTP over a single endpoint. Any client that supports remote MCP servers over HTTP can connect to it.
POST https://api.recordport.app/mcp
Authentication
Send your API key as a Bearer token. Organization and department keys are created in Console ▸ Integrations ▸ AI and API access; a member creates their own personal key in Console ▸ Account ▸ Connect your AI. Every key carries its own scopes, so a client can be given search and read and nothing else — two of the eight tools write, and they answer only for a key that holds the matching write scope. Treat keys like passwords and rotate any key you suspect has leaked.
Authorization: Bearer rp_your_key
Setup
Register the server once in your client and the tools appear automatically. Replace rp_YOUR_KEY with a key from the console.
Claude (claude.ai)
Settings ▸ Connectors ▸ Add custom connector, then enter the endpoint URL. When prompted for authentication, choose header-based auth and paste Bearer rp_YOUR_KEY as the Authorization header. Once connected, ask Claude things like "search our meetings about the Q3 launch" and it will call the tools below.
ChatGPT
Settings ▸ Connectors ▸ Advanced ▸ Add remote MCP server (developer mode), with the same endpoint and Authorization header. Availability depends on your ChatGPT plan.
Claude Code
claude mcp add voicecape --transport http \
https://api.recordport.app/mcp \
--header "Authorization: Bearer rp_YOUR_KEY"
Cursor
Add the server to ~/.cursor/mcp.json.
{
"mcpServers": {
"voicecape": {
"url": "https://api.recordport.app/mcp",
"headers": {
"Authorization": "Bearer rp_YOUR_KEY"
}
}
}
}
Other MCP clients
Any client that supports Streamable HTTP can connect with a configuration equivalent to this.
{
"type": "http",
"url": "https://api.recordport.app/mcp",
"headers": {
"Authorization": "Bearer rp_YOUR_KEY"
}
}
Tools
The server exposes eight tools over your organization's summaries, transcripts, decisions, and action items — six that read, and two that update the status of an action item or a decision.
| Tool | Arguments | Description |
|---|---|---|
| search_summaries | query, since, until, limit | Search your organization's summaries by keyword, optionally filtered by date range |
| get_summary | summary_id | Fetch one summary in full, with key points, decisions, and action items |
| get_transcript | summary_id | Fetch the speaker-labeled transcript of the recording behind a summary, with organization-assigned speaker names applied |
| ask | question | Ask a natural-language question and get an answer grounded in the most relevant summaries, with citations |
| list_action_items | done, assignee, since, until, limit | List action items across meetings with open/done status, assignee, and due date |
| list_decisions | status, since, until, limit | List decisions recorded across meetings, each with its review status and source summary |
| complete_action_item | id, done | Writes. Mark an action item done or reopen it, within the key's scope |
| set_decision_status | id, status | Writes. Change a decision's review status, within the key's scope |
Every key is scoped at creation: an organization-wide key reads across all departments, while a department-scoped key only reads that department's summaries. Every tool applies the scope automatically, and recordings a member keeps private are never exposed through API keys of either scope.
Example prompts
Once connected, your assistant decides when to call the tools. These prompts work well as starting points.
- "Search our meetings from last week and summarize what was decided about pricing."
- "List the open action items assigned to Minjun and draft a status check-in message."
- "Pull the transcript of the client onboarding call and extract every commitment we made."
Access and revocation
Keys follow the scope set at creation, either the entire organization or a single department. Within that scope a connected client can read summaries and transcripts, and can also update two things: whether an action item is done, and a decision's review status. Nothing else can be created, edited or deleted through the MCP server — recordings, summaries, members and settings are all read-only to it. Treat an MCP key as you would a teammate's account: give it to a client you trust with those two updates. You can revoke a key at any time in Console ▸ Integrations ▸ AI and API access, and a revoked key stops working immediately.
Prefer a plain HTTP export?
The same keys also work with the export API, a single GET request that returns your summaries as Markdown or JSON for scripts and scheduled jobs.
Create a key and connect your assistant
API keys live in Console ▸ Integrations ▸ AI and API access, ready in under a minute.
Start freeAlready have an account? Sign in