Rollbar MCP Server

The Rollbar MCP server connects an AI assistant to your error monitoring at Rollbar. Anyone who fixes production errors knows the routine: open the item in the dashboard, read the stack trace, check the affected version and environment, then look for the matching spot in the editor. The server shortens that path, because the assistant fetches the error data itself and can compare it directly with the code in your project. The server is maintained by Rollbar itself as an open repository and is licensed under MIT.

What is the Rollbar MCP server?

The server implements the Model Context Protocol (MCP), an open standard through which AI applications talk to tools. It runs locally as a process on your machine that the client starts itself via npx; the repository does not mention a hosted endpoint from Rollbar for this purpose. The server talks to the Rollbar API and passes the responses on to the assistant. The repository rollbar/rollbar-mcp-server is not archived and was last updated in October 2026.

Which tools the server provides

The README lists eleven tools. For orientation there are list-projects, get-top-items for the errors with the most occurrences in the last 24 hours, and list-items, which filters by status, severity, environment, or text. get-item-details returns an item with its latest occurrence in one call, list-occurrences shows individual occurrences page by page, and list-item-events the history with comments and status changes. For releases there are get-deployments and get-version, which shows how a version, such as a Git commit, behaved in an environment. get-replay loads data from a session replay. Only post-item-comment and update-item write; the latter changes status, severity, title, assignee, or team of an item.

Setup and tokens

You need a Rollbar account, Node.js with npx, and an MCP client such as Claude Code, Codex, or Cursor. In the configuration you enter npx -y @rollbar/mcp-server@latest as the command and set exactly one token variable: ROLLBAR_ACCESS_TOKEN for a single project or ROLLBAR_ACCOUNT_ACCESS_TOKEN for all projects of the account. For comments a user token in ROLLBAR_USER_ACCESS_TOKEN is also required. If two token types are set, the server does not start according to the README. Several projects can also be described in a file .rollbar-mcp.json that ROLLBAR_CONFIG_FILE points to. The default endpoint is https://api.rollbar.com/api/1 and can be replaced via ROLLBAR_API_BASE.

Security and privacy

Error reports often contain more than you expect: stack traces, variable values, request parameters, user identifiers. Everything the assistant retrieves enters the language model's context and thus reaches your client's model provider. Therefore issue tokens with read-only access where possible and limit them to the project you are working on. Only update-item and post-item-comment need write permissions; approve such calls only after your own review, since they change the handling state for the team. Keep tokens out of committed files.

Costs

The server is open source and free. According to Rollbar's pricing page there is a permanently free plan with limited occurrences and session replays per month and shorter retention, plus paid plans with more volume and longer retention and an Enterprise tier with custom terms. Current amounts and limits are on the official pricing page.

Who is the Rollbar MCP server for?

It is worthwhile for teams that already use Rollbar and want to fix errors together with a coding assistant: the assistant reads the item, finds the spot in the code, and proposes a fix, without you copying stack traces by hand. It also helps with the morning review, because get-top-items returns the most conspicuous errors. Anyone who does not use Rollbar gains nothing from it. Compared with copying from the dashboard, the advantage is that the assistant can ask targeted follow-up questions, for example about older occurrences or the behavior of a specific version.

A Rollbar account with an access token, Node.js with npx, and an MCP client such as Claude Code, Codex, or Cursor.

The Rollbar MCP server runs locally and gives your assistant access to errors, occurrences, and deployments from Rollbar. This guide walks through tokens, configuration, and the first calls.

Quick start

Register it as a stdio server with npx -y @rollbar/mcp-server@latest and set exactly one token variable (ROLLBAR_ACCESS_TOKEN or ROLLBAR_ACCOUNT_ACCESS_TOKEN).

npx -y @rollbar/mcp-server@latest

Step 1: Prerequisites

You need a Rollbar account, Node.js with npx, and an MCP client such as Claude Code, Codex, or Cursor.

Step 2: Choose a token

Create an access token in Rollbar. For a single project a read-only project token is enough; for several projects use an account token. Set only one of the two types, otherwise the server does not start. To change items the account token needs write permission, and to comment you need a user token.

Step 3: Register it in the client

In Claude Code, add an entry rollbar of type stdio to the file .mcp.json with the command npx and the arguments -y and @rollbar/mcp-server@latest, and set the variable ROLLBAR_ACCESS_TOKEN under env. In Cursor the JSON structure is the same; restart Cursor afterwards. For several projects, ROLLBAR_CONFIG_FILE points to a file .rollbar-mcp.json.

First calls

Ask the assistant to name the reachable projects with list-projects and the most frequent errors of the last 24 hours with get-top-items. Then have it explain an item with get-item-details and look for the matching spot in the code.

Use write permissions deliberately

update-item and post-item-comment change the state for the team. Approve such calls only after your own review and keep the token out of committed files.

Rollbar access token via environment variables: project token or account token (only one of the two); for comments additionally a user token.

Read projects, items, occurrences, deployments, and replays; with write permissions also update items and post comments.

The server runs locally and queries the Rollbar API. Retrieved error data enters the AI client context and therefore reaches the model provider.

Error reports can contain variable values, request parameters, and user identifiers; update-item and post-item-comment change the handling state for the team; tokens in configuration files can leak.