Jam MCP

AI agent debugging a Jam via MCP, with calls to getConsoleLogs, getNetworkRequests, and getUserEvents tools

A Jam records what happened on screen: the video, what the person said, every click, the console, and the network. Jam MCP gives your AI agent access to all of it. Paste a Jam link into Claude, ChatGPT, Claude Code, Cursor, or any MCP client, and the agent reads the recording itself. No one needs to retype what they saw or copy logs. All you need is a Jam link. Anyone can record the Jam: you, a member of your workspace, or a customer through a Recording Link. Jam MCP works with every Jam in your workspace, including ones recorded before you connected.

Connect your client

Jam MCP is one server at https://mcp.jam.dev/mcp. Most clients sign you in through your browser the first time you use a Jam tool.

Jam connector for Claude

Install the connector, sign in to Jam, and allow access to your workspace.

Add the server

Install the Jam plugin. It connects the MCP server and keeps itself up to date:

/plugin marketplace add jamdotdev/skills
/plugin install jam@jam-plugins

Or add the server on its own:

claude mcp add --transport http Jam https://mcp.jam.dev/mcp

Run /mcp, select Jam, and finish the sign-in in your browser. Claude Code also shows a notice at startup while a server still needs sign-in. For a session with no /mcp panel, such as claude -p or a CI job, sign in once from an interactive session first. If you added the server with claude mcp add, claude mcp login Jam does it from the shell. Or use a personal access token.

codex plugin marketplace add jamdotdev/skills
codex plugin add jam@jam-plugins

Start a new Codex session, then run /mcp to check the connection and sign in.

Open the command palette

Press Ctrl/Cmd + P and enter >MCP: Add Server.

Select the transport type

Select HTTP (HTTP or Server-Sent Events).

Enter the server URL

Enter https://mcp.jam.dev/mcp.

Name the server

Enter Jam.

Your mcp.json file should look like this:

{
  "servers": {
    "Jam": {
      "url": "https://mcp.jam.dev/mcp",
      "type": "http"
    }
  }
}

Gemini CLI

gemini extensions install https://github.com/jamdotdev/skills

Run /mcp auth Jam once to sign in.GitHub Copilot CLI

copilot plugin marketplace add jamdotdev/skills
copilot plugin install jam@jam-plugins

OpenCodeAdd the server to opencode.json, then run opencode mcp auth jam:

{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "jam": {
      "type": "remote",
      "url": "https://mcp.jam.dev/mcp"
    }
  }
}

Anything elseAny client that speaks MCP over HTTP works. Add https://mcp.jam.dev/mcp as a remote server and complete the sign-in when the client asks. If your client cannot open a browser, such as a CI job, sign in with a personal access token instead.

The plugin and the Jam skills are open source at github.com/jamdotdev/skills. To add the skills to any agent without the plugin, run npx skills add jamdotdev/skills.

Use it

Copy a Jam link

Copy a link from your workspace, a ticket, or a Slack thread. It looks like https://jam.dev/c/abc123.

Paste it and say what you want

One sentence is enough. The agent picks the tools, reads the Jam, and reports back.

Approve the tool calls

The first time, your client asks you to sign in to Jam and to allow the tools the agent calls. Approve them once and the agent keeps going.

What to say

Paste a Jam link and say what you want. Each guide in Jam for AI agents has the prompt to give and how to check the result. Start with Hand a Jam to your agent in a coding agent, or Turn Jams into tickets in Claude or ChatGPT.

To have the agent record proof of the fix, use the CLI. MCP reads Jams but cannot record one. The Jam CLI adds jam record, so a coding agent can record the fix running and hand back a second Jam link. See From report to test for the full loop.

Tools

Jam MCP exposes every read and write below. Your agent chooses among them. You only need this list to know what is possible, or to name a tool in a prompt.

Read a Jam

	Description
`getDetails`	Get a quick snapshot of the Jam: who made it, what happened, what it links to, and which other tools to try next.
`getConsoleLogs`	Grab the console logs from the Jam. Supports `logLevel`, `limit`, and `after` (page cursor) filters to reduce noise.
`getNetworkRequests`	List every web request from the Jam as JSON. Supports `statusCode`, `contentType`, `host`, `method`, `limit`, and `after` filters, plus `bodies` (`errors` by default, or `all`).
`getUserEvents`	Read each click, input, and page navigation change in plain language.
`getMetadata`	Read custom key-value metadata set via the `jam.metadata()` SDK: user IDs, app versions, feature flags, and more.
`getScreenshots`	Get every screenshot from a screenshot Jam for visual inspection.
`getVideoTranscript`	Get the spoken transcript from a video Jam where the microphone was enabled, in WebVTT format with timestamps.
`getVideoChapters`	Get AI-generated chapter markers for a video Jam: titled segments with start and end timestamps. Text only.
`getFrames`	See what was on screen in a video Jam. Return frames at specific timestamps, evenly across a window, or one labeled overview grid spanning the whole recording.

The video tools (getVideoTranscript, getVideoChapters, and getFrames) need a video Jam, and getScreenshots needs a screenshot Jam. Instant Replay Jams return nothing from those four. Every other tool works on every Jam type.

See what was on screen

getVideoTranscript returns what the reporter said. getFrames returns images from the recording so your agent can inspect the screen at the moments that matter. Call it three ways:

Overview grid. Set overview: true to get one image: frames evenly spaced across the whole video, stitched into a labeled grid where each cell shows its timestamp. The frame count scales with the video’s length (six for short clips up to 16 for long ones). This is the fastest way for an agent to orient itself on a visual bug before drilling in.
Specific timestamps. Pass at with one or more times in milliseconds to capture those exact moments.
Window. Pass count to sample evenly-spaced frames. Add fromMs and toMs to limit the range. Without them, the frames span the whole video from the first frame to the last.

Use size (small, medium, or large, default medium) to trade detail against your agent’s context budget. The overview grid ignores size and always renders small cells. Each frame comes back as an image preceded by a t=<ms> text anchor so the agent knows which moment it’s looking at.

getFrames needs a video that has finished processing, which takes a few minutes after upload. For a screenshot Jam, a video still processing, or an older video that has no frames, it returns a short text block that says why.

Search and browse

	Description
`searchJams`	Search Jams by text across titles, descriptions, comments, linked issues, and transcripts. Filter by author, folder, type, date, and more.
`listMembers`	Browse your team members.
`listFolders`	Browse available folders.

Search filters go well beyond author and folder. Your agent can find Jams by:

Page URL: Jams captured on a page URL such as /checkout.
Source: Jams from the extension, the Mac app, iOS, Recording Links, and more. Narrow further to a specific Recording Link, or to links that share a reference.
Linked issue: Jams sent to an integration, or with an issue key such as JIRA-123.
Environment: country, browser, OS, incognito session, etc.
Date: Jams created in a date range, such as the last seven days.
Exclusions: everything except a specific author, folder, or Jam type.

Combine any of these with search text in one request.

Comment and react

	Description
`createComment`	Add Markdown comments to a Jam, optionally mentioning a member to notify them.
`editComment`	Rewrite a comment you left earlier, so an agent can correct its own note instead of stacking a follow-up.
`deleteComment`	Delete a comment you wrote, along with its attachments.
`addReaction`	React to a comment. Handy for acknowledging one you’ve already acted on.
`removeReaction`	Take back a reaction you left on a comment.

Comments your agent posts show a label with the agent’s name, such as Claude Code or Cursor, so readers can tell them apart from comments people wrote. Jam asks agents to post one comment when they find the cause of a problem or fix it, and to mention the Jam’s creator so they get notified.

Organize Jams and folders

	Description
`updateJam`	Rename a Jam, rewrite its description, or move it between folders using a folder ID or folder name.
`addResource`	Attach a URL, such as the pull request or ticket a Jam proves, with an optional label.
`removeResource`	Remove a URL from a Jam’s resources.
`markRelated`	Mark a Jam as related to another Jam of your team: it fixes, proves, or relates to it. getDetails on the other Jam returns it.
`markUnrelated`	Remove every link from a Jam to another Jam, whatever its type.
`deleteJam`	Delete a Jam. It leaves your lists and search, and there is no way to restore it.
`createFolder`	Create a folder to organize Jams into. Returns the folder’s ID and short ID so a follow-up `updateJam` can file a Jam into it.
`updateFolder`	Rename a folder, addressed by its name, ID, or short ID.
`deleteFolder`	Delete a folder and every Jam inside it, none of them restorable. Returns how many Jams went with it.

Links show in the Jam’s Info under Resources and Related Jams, and in its timeline with the agent’s name. A link goes both ways: when a fix Jam is marked as the fix for a bug Jam, the bug Jam lists it as fixed by. Jam asks agents to attach the pull request, ticket, CI run, or doc a Jam belongs to, and to mark a Jam that proves a fix as the fix or the proof for the bug Jam. Resource URLs show on the Jam’s share page, which can be public. Jam refuses a URL with a username or password, and asks agents never to attach one that carries a token.

Recording Links

	Description
`listRecordingLinks`	List the workspace’s Recording Links, newest first, with cursor pagination and an `includeRevoked` option.
`getRecordingLink`	Get one Recording Link by its public ID.
`listRecordingLinkJams`	List the Jams recorded through a link, by public ID, with cursor pagination.
`createRecordingLink`	Create a reusable Recording Link, optionally targeting a folder, connected domain, or expiration.
`updateRecordingLink`	Edit a link’s name, folder, reference, Jam title, expiration, or metadata by public ID (revoked links can’t be edited).
`deleteRecordingLink`	Revoke a link by its public ID (soft-delete, so collected Jams stay).
`listRecordingUrls`	List the team’s connected recording domains.
`getRecordingUrlVerifyLink`	Start verifying a connected domain and return a link a human opens to confirm it.

Send feedback

	Description
`submitFeedback`	Report a tool, docs page, or skill that was wrong, stale, or confusing. Agents send it themselves, and Jam’s team reads every submission.

Every tool has a matching Jam CLI command that returns the same fields with --json, so a workflow you prototype in chat runs unchanged in a script.

Manage access

Go to Settings → MCP to see every connected client. Select one to revoke its access at any time. Personal access tokens live on the same page.

MCP mirrors your existing Jam permissions. It does not grant new access beyond what a user can already see in the Jam web or mobile apps. Use your normal admin controls for members and roles, or SSO in Settings.

FAQ

Do I need the Chrome extension?

No. You only need a Jam link. Anyone can record the Jam: you, a member of your workspace, or a customer through a Recording Link. To record Jams yourself, install the Chrome extension, Jam for Mac, or the iOS app.

Can I paste multiple Jam links at once?

Yes. One Jam at a time keeps the agent inside its context window. For a batch, ask for one summary per Jam first, then reason across the summaries.

Does Jam MCP send my Jam data to third parties?

The MCP server returns your Jam’s data to the client you connected, and nowhere else. Some fields it returns, such as transcripts, chapters, and AI-written descriptions, were generated when the Jam was processed, under Jam’s AI policy. Jam opts out of training on customer data and takes steps to de-identify it. See the Security page for details.

Does MCP work with Instant Replay Jams?

Mostly. The video tools (getVideoTranscript, getVideoChapters, and getFrames) and getScreenshots return nothing for Instant Replay Jams. Console logs, network requests, user events, metadata, search, comments, and every other tool work the same.

Can I use MCP in CI or without a browser?

Yes. Sign in with a personal access token in an Authorization: Bearer header. For CI jobs and scripts, the Jam CLI is usually the better fit: it runs anywhere a shell runs and can also record Jams.

MCP

CLI

SDK

Webhooks

Connect your client

Jam connector for Claude

Jam app for ChatGPT

Jam plugin for Cursor

Use it

What to say

Tools

Read a Jam

See what was on screen

Search and browse

Comment and react

Organize Jams and folders

Recording Links

Send feedback

Manage access

FAQ

​Connect your client

Jam connector for Claude

Jam app for ChatGPT

Jam plugin for Cursor

​Use it

​What to say

​Tools

​Read a Jam

​See what was on screen

​Search and browse

​Comment and react

​Organize Jams and folders

​Recording Links

​Send feedback

​Manage access

​FAQ

Connect your client

Use it

What to say

Tools

Read a Jam

See what was on screen

Search and browse

Comment and react

Organize Jams and folders

Recording Links

Send feedback

Manage access

FAQ