Yes, there is an MCP server for Azure DevOps sprint and flow metrics. The Agile Analytics local AI connector runs on your own computer, signs in as you, and lets Claude, Claude Code, VS Code or any assistant that can run a local MCP server ask for a sprint summary, cycle time, flow efficiency, work in progress, aging and a forecast. It answers with computed numbers: counts, dates and work item ids only, never titles or names.
What the connector is
MCP, the Model Context Protocol, is how an AI assistant calls tools on your machine. The connector is an npm package, @baytekdev/agile-analytics-mcp, that your assistant starts with npx. It is part of your Agile Analytics subscription: no extra cost and no licence step. You need:
- Agile Analytics installed in your Azure DevOps organisation, version 6.75.1 or later.
- A saved workflow mapping for the project. The connector never guesses a mapping.
- Node.js on your computer (npx comes with it).
- An assistant that supports local MCP servers, such as Claude Desktop, Claude Code or VS Code.
How it differs from a general-purpose Azure DevOps MCP server
General-purpose Azure DevOps MCP servers give your assistant raw work items and other project data, and the assistant works out any metric itself. That is useful for many jobs, but a model counting items does not know which states your team treats as active or done, so its cycle time can be wrong.
- Computed metrics, not raw items. The connector returns the same numbers the Agile Analytics views show, worked out with the same metric code and your saved workflow mapping.
- Less content leaves Azure DevOps. Your assistant gets counts, dates and work item ids. Titles, descriptions, comments, tags and assignees are never read.
- Read only. It never changes anything in Azure DevOps.
The six tools
Sprint summary, cycle time and flow efficiency use finished sprints only, the same rule the views follow. Work in progress and aging look at the running sprint.
| Tool | What it returns |
|---|---|
sprint_summary | One finished sprint: delivered items and points, cycle time, carryover, commitment, reliability (say/do) and scope added and removed. |
cycle_time | Count, average and p50/p70/p85/p95 days over the last 3, 6 or 12 finished sprints, with the ids of the slowest items. |
flow_efficiency | Active time against waiting time over the last 3, 6 or 12 finished sprints. |
wip | Items in progress right now in the running sprint, by workflow stage, against your saved WIP limit. |
aging | Open work in the current sprint, carried-over items included: id, stage, days in that stage and days in flow. |
forecast | A Monte Carlo forecast from the last 12 sprints of throughput: how many items in a number of days, or when a backlog of a given size finishes. |
Set it up in three steps
Set-up takes three steps. The guide has the full detail for each assistant and operating system.
- Create a read-only token. In Azure DevOps, open User settings → Personal access tokens → New Token and give it exactly three scopes: Work Items (Read), Project and Team (Read) and Extension Data (Read). On Windows, save it as a user environment variable named
AZURE_DEVOPS_PAT; on macOS and Linux, the guide shows where it goes instead. Organisations connected to Microsoft Entra ID can runaz logininstead. - Add the connector to your assistant. In Claude Code, one line does it:
claude mcp add agile-analytics -- npx -y @baytekdev/agile-analytics-mcp. In Claude Desktop, open Settings → Developer → Edit Config. In VS Code, use.vscode/mcp.json. The guide has the snippet for each. - Restart your assistant and ask. Name the organisation, project and team in your question.
Related: Set-up guide: the local AI connector
How to ask Claude, Claude Code or VS Code about sprint metrics
Once the connector is added, ask in plain words. In VS Code, MCP tools are available in agent mode. Your assistant picks the tool and gets the numbers back; you can then ask it to explain them, compare sprints or draft notes for a sprint review.
- "Sprint summary for team Platform in project Web Store, org contoso."
- "Cycle time over the last 6 sprints for team Platform in project Web Store, org contoso."
- "Which items are aging for team Platform in project Web Store, org contoso?"
- "How many items can team Platform in project Web Store, org contoso, finish in the next 30 days?"
Privacy: what it reads and what it sends
The connector uses your own Azure DevOps access, which already reaches the same data, and does not widen it.
- Runs on your computer. The connector talks only to
dev.azure.comandextmgmt.dev.azure.com, as you, with read requests. Nothing is written to your disk, and the token is kept in memory only. - Reads what the metrics need. State, type, dates, sprint and area path, estimates, board column, the blocked flag and their history, plus your team settings and your Agile Analytics settings.
- Sends nothing to Baytek. There is no licence check and no telemetry.
- Your assistant's provider. What your assistant does with the numbers is between you and your assistant's provider, so check that provider's terms as you would for any prompt.
What admins control
An organisation admin can refuse every connection in Configuration → General with the "Allow the local AI connector" switch. It is on by default, and only Agile Analytics admins can change it. The connector checks the switch before every answer and replies "disabled by admin" when it is off. Each change is recorded in the Admin activity log.
Try it
The connector is part of every plan and is available during the trial. The 30-day free trial starts on install with no credit card, then it is from $300 a year per organisation, with every feature on every plan.
Related: Azure DevOps flow metrics · Install instructions · Pricing