AI Usage Analytics with GitHub Copilot + One-Click MCP Connection đ¤đ
We are thrilled to announce the AI Usage dataset, starting with GitHub Copilot metrics: you can now measure how your engineers adopt AI coding assistants, right next to the pull requests they ship.
Alongside this, connecting your AI assistant to the Keypup MCP server now takes a single click, and the AI Agent is out of beta.
AI Usage Dataset: Measure AI Adoption Across Your Team
Your engineers use AI coding assistants every day, but how much, on which tools, and with what result? Until now, that answer lived in each vendor's admin console, disconnected from your delivery data. To solve this, we are introducing the AI Usage dataset!
- Daily Usage Metrics: Interactions, suggestion acceptance, credits consumed and code contribution, broken down by user, model, surface (IDE completion, chat, agent mode, CLI, code review), client and language.
- GitHub Copilot, with One Year of History: The GitHub integration now retrieves Copilot usage metrics. Enable the Copilot Metrics project that appears in your list of GitHub projects, and Keypup imports up to one year of history on the first import.
- The AI Adoption & Usage Dashboard: A new dashboard template tracks active developers, adoption by model and surface, suggestion acceptance rate, credit consumption, code contribution, a developer leaderboard and pull request outcomes.
- Everywhere You Report: The dataset is available in insights, the API and the MCP server, and both the AI Agent and the MCP server know how to query it. Try asking your AI assistant "Which models do our engineers use most, and has our acceptance rate improved this quarter?"
GitHub Copilot is only the start. The AI Usage dataset is built to hold every AI coding assistant side by side, so you can compare them on the same insights. Anthropic Claude (including Claude Code) is coming next, followed by Cursor and OpenAI (the API Platform and Codex). Stay tuned! đ¤
Every field is documented in the AI Usage dataset collection. To get started, check out our GitHub Copilot integration article for the prerequisites.
MCP Server: Connect in One Click with OAuth
Connecting an AI assistant to the MCP server used to mean generating an API token and pasting it into a configuration file. The MCP server now supports OAuth: your AI client opens Keypup, you approve the connection, and you're done.
- Simpler and More Secure: OAuth tokens expire and are never handled by hand, so there is no long-lived token to store or leak. API tokens remain available if you prefer them.
- Automatic Client Registration: Clients that support Client ID Metadata Documents, such as VS Code and Claude, register themselves automatically. We have also pre-registered several popular clients and will keep adding more.
- MCP Setup Page: A new MCP Setup page, available from the sidebar, gives the setup instructions for each AI client, including ChatGPT, with screenshots.
- Clear Scopes: The new
mcp:readandmcp:writescopes enable every MCP capability, now and in the future, and are the default when you connect over OAuth. Every scope now has an info bubble explaining what it grants, and new scopes let you manage Pin Groups through the API.
The MCP server has also learned a few things: it can read your team's AI Instructions, and the list of teams now includes their connected apps, so your AI assistant starts every conversation with the right context. Check out the updated Using the MCP server documentation.
Other improvements and bug fixes
- Improvement: AI Agent: The AI Agent is out of beta, and now runs on Gemini 3.8 Flash by default.
- Improvement: AI Agent: Custom fields are no longer all sent to the AI upfront. The AI Agent now looks them up when your question needs one, which makes answers faster and fixes a permanent error for teams with a very large number of custom fields.
- Improvement: Performance: Insights that filter on custom fields are faster, when using the "equal" and "contains" operators.
- Improvement: Jira: Changes to Fix Versions, such as a new release date, are now reflected on the associated issues in near real time. Previously they could get out of sync, because Jira does not send an issue update when a Fix Version changes.
- Improvement: Pin Groups: The open or closed state of each pin group on the sidebar is now remembered across page reloads.
- Improvement: MCP Server: When a team has no active plan, the MCP server now responds with an explanation and a link to the billing page, instead of a generic error.
- Bugfix: Bitbucket: Bitbucket restricted access to review data without notice. Bitbucket no longer lets us retrieve the number of approvals required on a pull request. A ticket has been opened with them to reinstate this feature.
- Bugfix: Business Duration Operators: Fixed a calculation that added one day to BUSINESS_DAYS durations when the start date's time of day was later than the end date's.
- Bugfix: AI Agent: The AI Agent can now add OR and AND filter groups on every type of insight, not only card feeds.
- Bugfix: Custom Formulas: An invalid custom formula now returns a detailed error explaining what is wrong, instead of a generic error.
- Bugfix: Custom Fields: Fixed the names of Jira custom fields built from lists of values, such as Fix Version release dates.
- Bugfix: Pin Groups: Fixed the "others" group staying collapsed after switching teams, which made the list of dashboards look empty.
- Bugfix: Dashboards: Choosing "edit" on a dashboard from the collection view now opens it in edit mode, and saving no longer sends you back to the collection.
- Bugfix: Login: Fixed an issue where you could land on the onboarding page instead of your dashboards after logging in, and signup and login links now keep their parameters.
- Bugfix: MCP Server: Fixed a response format issue that prevented some AI clients from listing teams.














