How to Connect the Sentry MCP Server and Triage New Errors
Add Sentry's hosted MCP server as a custom server with one-click sign-in, then build a workflow that triages new issues and posts a digest to Slack.
What This Integration Does
Sentry hosts an MCP server for working with your organizations, projects and issues. Spojit has no built-in Sentry connector, so you add it as a custom MCP server. Sentry supports automatic sign-in, which means you paste a URL, leave authentication on Auto, and approve access in Sentry. There is no API key to create and no OAuth app to register, which is how most hosted MCP servers connect to Spojit.
The workflow runs on a schedule. It asks an agent to look at unresolved issues in a project, group them by what is actually breaking rather than by stack frame, and write a short triage digest. Nothing is written back to Sentry, so the run is read-only and safe to schedule aggressively while you tune the prompt. Each run is independent: it reports the current state rather than tracking what it said last time.
Prerequisites
- A Sentry account with access to the organization you want to connect. The tools can only see what your Sentry user can see.
- Your Sentry organization slug, and the project slug if you want to scope the connection to one project. Both appear in your Sentry URLs, for example
https://<organization-slug>.sentry.io/projects/<project-slug>/. - Permission to add connections in Spojit.
- A Slack connection, if you want the digest posted to a channel.
Step 1: Pick Your Server URL
Sentry recommends scoping the connection as narrowly as you can. A scoped URL defaults the tools to that organization or project and hides the discovery tools you no longer need, which keeps agent runs focused and cheaper:
- Everything your user can access:
https://mcp.sentry.dev/mcp - One organization:
https://mcp.sentry.dev/mcp/<organization-slug> - One project:
https://mcp.sentry.dev/mcp/<organization-slug>/<project-slug>
For a triage workflow on a single service, use the project-scoped URL.
Step 2: Add the Custom Server in Spojit
Go to Connections and click Add custom server. Enter:
- Name:
Sentry, or the project name if you plan to connect several. - Server URL: the URL you chose in Step 1.
- Authentication: leave it on Auto.
Choose a visibility, Organization so everyone in the workspace can use it or Private for just you and admins, then click Save & connect.
Step 3: Approve Access in Sentry
Sentry opens its consent page. Sign in, review the access being requested, and approve. Spojit stores the tokens, marks the connection Connected, and refreshes them automatically from then on. This is the whole of Auto mode: Spojit asks the server how to sign in, registers itself with Sentry's login service, and sends you there to approve.
Step 4: Add Sentry to a Workflow
Create a workflow with a Schedule trigger set to the time you want the digest, for example every weekday at 8:30am. Add a Connector node and open Select Connector: custom servers appear in their own Custom Servers section at the top of the picker. Select your Sentry server and set the node to Agent Mode.
Step 5: Write the Triage Prompt
Paste this into the Prompt field:
You have Sentry tools available for this project.
1. List the unresolved issues from the last 24 hours.
2. Group them by root cause, not by stack frame: issues with the
same underlying failure belong together even if the messages differ.
3. For each group give: a one-line description in plain language,
the number of events, the number of users affected, and whether
it is new or a regression.
4. Rank the groups by users affected, highest first.
5. If nothing is unresolved in that window, say so in one line.
Keep it under 200 words.Narrow Available Tools to the issue and event tools so the agent does not spend turns on discovery.
Step 6: Deliver the Digest
Add a slack connector node after the Sentry node in Direct Mode, choose the post-message tool, and reference the agent output with {{ step1 }}. Direct Mode is deliberate here: the reasoning already happened in the Sentry node, so there is no reason to pay for a second agent just to send a message. A Send Email node works the same way if your team prefers email.
Tips
- Scope the URL to a project per workflow rather than connecting once at organization level and filtering in the prompt. It is cheaper and the agent makes fewer wrong turns.
- Run it manually a few times before scheduling. Triage prompts need tuning against your actual error mix.
- Keep the run read-only while you tune. Ask for a report, not for issues to be resolved or assigned.
- Re-saving the connection refreshes the tool list, so tools Sentry adds later become available without recreating it.
Common Pitfalls
- Tools asking for an organization every time: the connection is using the unscoped URL. Edit the connection, change the Server URL to the scoped form, and save to reconnect.
- A project is missing: the tools act as your Sentry user. Confirm that user is a member of the team that owns the project.
- Expecting the digest to remember yesterday: each run is independent. If you want "what changed since the last digest", write the previous summary to your file store and feed it into the next run.
- Noisy first runs: a project with a long backlog of unresolved issues produces a wall of text. Narrow the time window in the prompt before you widen it.
Testing
Run the workflow manually before putting it on the schedule. Check the execution log for the Sentry node: the tool calls it chose tell you whether the scoped URL is working, and the output tells you whether the prompt is producing something a human would actually read at 8:30am. If the node fails to authenticate, open Connections and use Reconnect to re-run the sign-in.