---
title: MCP server | ReplayPilot Docs
description: Let an AI agent query your sessions directly over the Model Context Protocol.
canonical: https://replaypilot.com/docs/mcp-server
---

# MCP server | ReplayPilot Docs

Search docs…⌘KBrowse docs

Get started
- [Installation](/docs/getting-started)
- [What's new](/docs/whats-new)

Dashboard
- [Replays & the session player](/docs/replays)
- [Errors](/docs/errors)
- [Frustration signals](/docs/frustration)
- [Funnels](/docs/funnels)
- [Releases](/docs/releases)
- [Logs & network requests](/docs/logs)
- [Visitors](/docs/visitors)
- [Heatmaps](/docs/heatmaps)
- [Click map](/docs/click-map)
- [Cohort retention](/docs/cohorts)
- [Projects & project keys](/docs/projects)
- [Alerts](/docs/alerts)
- [Shared links](/docs/shared-links)

Integrations
- [Integrations](/docs/integrations)
- [MCP server](/docs/mcp-server)

Account & data
- [Billing & plans](/docs/billing)
- [Security & masking](/docs/security-masking)
- [Data retention & limits](/docs/retention-limits)
- [Team accounts](/docs/team-accounts)

# MCP server

Let an AI agent query your sessions directly over the Model Context Protocol.

## What MCP is

MCP (Model Context Protocol) is a standard way for an AI agent or tool, like Claude or another
assistant, to call directly into an app. No copying data back and forth by hand. ReplayPilot runs
its own MCP server, so an agent can query your sessions, errors, and logs on its own and use what
it finds to help you debug.

This is separate from the small AI assistant built into the ReplayPilot marketing site. The MCP
server is for external AI tools you connect yourself, using your own API key.

## Server address

Point your AI tool at:

https://mcp.replaypilot.com/mcp

It speaks the Streamable HTTP transport. This is the address ReplayPilot gives every customer. Check your AI tool's own docs for how it
wants the address entered: the field name varies by tool.

## Authentication

Every request needs a Bearer token, using one of two keys:

- **Project secret key**: from that project's Keys tab. See [Projects & project keys](/docs/projects). This scopes the AI agent to that one project only.

- **Account master key**: from Settings → Security → MCP. This scopes the AI agent to your whole account. Because a master key can reach any of your projects, every tool call the agent makes needs to say which project it means.

Without a valid key, the server still answers a basic "what tools exist" request, so an AI tool
can see what's available before you've connected anything. But every real tool call (asking for
session data) replies asking you to get a key from Settings first.

## What an AI agent can do

Every tool the server offers reads data. None of them create, change, or delete anything in
your account.

- List sessions, with filters for errors only, frustration only, browser, and minimum duration.

- Get full detail on one session, including its action log.

- List error groups.

- List frustration groups (rage clicks and dead clicks).

- List log groups (console and network).

With an account master key, it can also list every project on your account, so it knows which
project id to pass to the other tools.

Every list tool returns a nextCursor. Pass it back as the cursor argument to get the next
page; it's null once there's nothing left.

## Limits

- 60 requests per minute per IP address: the same limit as ReplayPilot's session-ingest endpoint.

- A list tool returns at most 100 items per page.

- Error and frustration group listings look at your project's most recent 1,000 sessions. You
can page through that window, but not past it.

These limits keep responses fast. If an AI agent needs something older than that window, ask it
to narrow the request, for example with a browser or page filter.

[← Integrations](/docs/integrations)[Billing & plans →](/docs/billing)