Microsoft Scope
Audit or test tool operated by Microsoft.
In short
An open-source research-preview platform that evaluates how AI agents use a product on tasks you define, with reusable success criteria and the evidence behind every result. It ranks 9 of 10 on the ten-criterion rubric with 27 of 100, earning most on real agents 12 of 15 and dev fit 7 of 10.
What it is
An open-source research-preview platform that evaluates how AI agents use a product on tasks you define, with reusable success criteria and the evidence behind every result.
Operator: Microsoft. Type: Audit or test tool. Home page: microsoft.github.io/scope.
Ten-criterion rubric
| Criterion | Points | What the source says |
|---|---|---|
| Journey: find | 0 of 4 | none |
| Journey: read | 0 of 4 | none |
| Journey: use | 4 of 4 | agents attempt tasks against the product 1 |
| Journey: pay | 0 of 3 | none |
| Checks | 2 of 15 | not a check scanner 1 |
| Real agents | 12 of 15 | runs coding agents on repeatable tasks; access is requested 1 |
| Open method: catalog | 0 of 3 | no site checks |
| Open method: weights | 0 of 3 | none |
| Open method: scale | 0 of 2 | none |
| Open method: changes | 0 of 2 | the repository has no releases and no change log 2 |
| Standards: cites | 0 of 3 | no fixed check catalog: criteria are the user's own, and its docs map none to a spec (0 percent) 3 |
| Standards: maintains | 0 of 2 | none |
| Dev fit: api | 3 of 3 | generated API reference 1 |
| Dev fit: mcp | 0 of 3 | the documentation covers importing MCP servers for the agents under test; Scope does not expose one 3 |
| Dev fit: cli | 2 of 2 | its README says the evidence behind every result can be inspected through the Portal or the CLI 1 |
| Dev fit: ci | 2 of 2 | benchmarks run inside a development workflow 1 |
| Peers: ranking | 0 of 6 | no score or leaderboard 1 |
| Peers: size | 0 of 4 | none |
| Fixes: guidance | 0 of 4 | none |
| Fixes: brief | 0 of 3 | none |
| Fixes: applies | 0 of 3 | none |
| Repeat scans | 1 of 5 | versioned agent profiles so a later run starts from a known setup; no data 1 |
| Free: free | 0 of 3 | access is requested 1 |
| Free: limits | 1 of 2 | MIT-licensed code on GitHub 1 |
Total 27 of 100, rank 9 of 10. The ranking is on the pick page; the same cells are on the evidence page.
What the sources say
- Microsoft describes Scope as an open-source agentic experience evaluation platform; the repository description adds that it is a research preview.1
- It evaluates how AI agents use a product, approach real tasks and respond to feedback; success is defined with reusable criteria, each run can be followed live, and the evidence behind every result is inspected through a Portal or a CLI.1
- The repository is MIT licensed, was created on 2026-07-28, was last pushed on 2026-11-04 and had 75 stars on 2026-11-05.4
Pricing and access
Open source under the MIT licence; no price is published (read 2026-11-05). 1
What it does not do
- Its README describes no public score for sites and no leaderboard.
- Evaluates what agents do on tasks you define; the README mentions no static checks of files such as llms.txt or robots.txt.
Where to read next
Shown next to Ora because Ora's Journey also points a real agent at a domain with one task; Ora's scan runs no agent, and Scope has no public score.
Sources
- github.com/microsoft/scope. https://github.com/microsoft/scope, read 2026-11-05.
- github.com/microsoft/scope/releases. https://github.com/microsoft/scope/releases, read 2026-11-05.
- microsoft.github.io/scope. https://microsoft.github.io/scope/, read 2026-11-05.
- api.github.com. https://api.github.com/repos/microsoft/scope, read 2026-11-05.
Last reviewed 2026-11-05