# Session replay MCP servers: give your coding agent the replay

**Short answer:** several replay tools now let a coding agent read recordings over MCP. PostHog, OpenReplay, Rollbar and UserTapes expose real-user sessions; Replay.io exposes recordings you make in its browser, CLI or Playwright. BugTape is different in scope: it does not record every visit. It keeps a short DOM replay around each error, failed request or user report, groups repeats into one issue and gives your agent that evidence over hosted MCP. Pick by what you want the agent to read.

No account: `npx -y bugtape --help` prints the commands and does not sign you in. The demo MCP at https://app.bugtape.ai/demo/mcp needs no sign-in. Your own workspace needs one.

Page owner: Jaco Veldsman, hi@bugtape.ai. Last checked 2026-11-02. I build BugTape. Each vendor row is from that vendor's own docs on 2 November 2026. This page quotes no vendor prices; read them on each vendor's pricing page.

## Session replay MCP servers compared

| Tool | What the agent reads | Recordings come from | MCP setup | Source |
|---|---|---|---|---|
| PostHog | Search and read session recordings, playlists | Real users (PostHog's web SDK) | `npx @posthog/wizard mcp add`; Cursor, Claude Code, Codex, Windsurf, VS Code | [PostHog docs](https://posthog.com/docs/session-replay/surfaces/mcp) |
| OpenReplay | Sessions replayed in chat with DOM reconstruction; mobile sessions get a link | Real users, OpenReplay Cloud or self-hosted | Local stdio app with browser login; Claude Desktop, Claude Code, ChatGPT, Codex | [OpenReplay MCP docs](https://docs.openreplay.com/en/mcp/) |
| Rollbar | Errors, stack traces, deploys and session replays (`get-replay`) | Real users (Rollbar SDK) | `npx -y @rollbar/mcp-server@latest`, local stdio, MIT | [Rollbar MCP](https://rollbar.com/mcp-server) |
| UserTapes | Tapes filtered by errors or rage clicks, read as a markdown transcript (11 tools) | Real users | Hosted at `https://usertapes.com/mcp`, read-scoped token separate from ingest; Claude Code, Claude Desktop, Cursor | [UserTapes MCP docs](https://usertapes.com/docs/mcp) |
| Replay.io | Time-travel debugging: console, source, network, React, Playwright | Recordings you make with the Replay browser, CLI or Playwright | Hosted at `https://dispatch.replay.io/nut/mcp`; public recordings need no key | [Replay MCP docs](https://docs.replay.io/basics/replay-mcp/overview) |
| BugTape | DOM replay, console, network and stack around one failure (`get_repro_context`, `get_bug_events`) | Real users, only around errors, failed or hung requests and user reports | Hosted at `https://app.bugtape.ai/mcp`, OAuth or `bt_pat_` | [MCP tools](/docs/mcp-tools/) |

## What BugTape does

The browser SDK keeps a short buffer and sends it with an uncaught error, a failed or hung request, or a user's report. Repeats of the same failure become one issue. DOM replay is rebuilt from events; it is not a screen video. BugTape is not product analytics and has no heatmaps.

Your agent connects once. `list_recommended_critical` is the ranked inbox. `get_repro_context` is the agent-first pack: bug metadata plus a compact console, network and error timeline. `get_bug_events` returns the replay events themselves (DOM, network, console, errors), filterable by type. `ack_bug` with `working`, then with `prUrl`, shows your team Picked up and PR opened. A human reviews the pull request. When you approve the fix it is anchored to that release, and if real users hit the same failure on a later release BugTape reopens it as a regression.

## How to choose

- **You already run PostHog, OpenReplay or Rollbar:** use its MCP server. The agent reads the replays you already have.
- **You want the agent to step through one recording in depth:** Replay.io's time-travel debugger, on recordings you make.
- **You want rage clicks and UX friction as text:** UserTapes.
- **You want the agent to read only the sessions where something broke, grouped into issues it can acknowledge:** BugTape.

## FAQ

### What is a session replay MCP server?

An MCP server that lets a coding agent such as Claude Code, Cursor or Codex search or read recorded user sessions, so it can see what a user did before a bug.

### Does BugTape record every session?

No. BugTape sends a short buffer only with an error, a failed or hung request, or a user report. It is not a full-session recorder or an analytics tool.

### Is BugTape's replay a video?

No. DOM replay is rebuilt from events. It is not a screen video.

### Does BugTape replace Sentry Replay or PostHog?

No. Keep them if you rely on full-session replay or analytics. BugTape fits beside them for failures handed to your agent.

### Can I try a session replay MCP without an account?

BugTape's demo MCP at https://app.bugtape.ai/demo/mcp is read-only and needs no sign-in.

## See also

[DOM replay vs screen video](/blog/dom-replay-vs-screen-video/) · [Best error tracking MCP servers](/blog/best-error-tracking-mcp-servers/) · [BugTape vs Rollbar](/compare/rollbar/) · [LogRocket alternatives](/alternatives/logrocket/) · [Compare](/compare/)

## Try the commands

```bash
npx -y bugtape connect claude   # or cursor, codex
```

Then click Allow once. [Start free](/console/register?utm_source=bugtape_seo&utm_campaign=session-replay-mcp). No card.

Your team reviews the PR. Agents do not silent-ship.