Decide which scopes your AI agents get.

Approve OAuth scopes and MCP tool access before your agents ship, with a record of who decided and why.

Explore the demo

Critical

A read-only tool holds a scope that can change data

The MCP tool lookup_customer says it only reads, yet holds crm:write.

ScopeNeat recommendation

Give the tool only read scopes.

Sources

OWASP Top 10 for LLM 2026 LLM03:2026, mitigation 4. Published security guidance

Supports evidence for

RTS 2024/1774 Art. 21(a), ISO/IEC 27001 A.8.2

What this means

Accountable MCP server owner

Know how binding each finding is.

A MUST never hides behind a blog post.

Defined by the standard: 6
Recommended by the standard (SHOULD): 5
Published security guidance: 7
YogaSec practice: 10

28 rules, each with a quoted source.

Put your approval workflow into action.

Segregation of duties, enforced at every step.

YogaSec's reading of DORA Art. 9(4)(e): evidence, not compliance.

Tested

Stop unapproved changes at the pull request.

$ scopeneat scan --baseline baseline.json

Compared with the approved baseline:
  BLOCKING  New tool: wire_money

Failed: 1 change(s) go beyond the approved baseline.

Demo data, output shortened.

Hand your auditors evidence by requirement.

# ScopeNeat findings by requirement

## DORA Art. 9(4)(c), 8 findings
## DORA Art. 9(4)(d), 1 finding
## RTS 2024/1774 Art. 21(a), 13 findings
## RTS 2024/1774 Art. 21(b), 1 finding
… 11 more requirements

Demo data, output shortened. A finding supports evidence for these requirements. On its own it does not make an organization compliant.

Works beside your security tools.

It doesn't scan servers or block traffic.
It decides which scopes agents get, records why, and gates changes in CI.
Your IAM team applies the approved decision.

Explore the demo workflow.

Five roles, one request, the full analysis engine. Example data, no account, nothing saved.

Explore the demo Questions and answers