Authentication and authorization
Review server access, token audience and scope, delegated authority, and whether permissions are enforced for every operation.
BASTET / SERVICES
MCP Security Assessment
Review the trust boundary between an AI agent and the systems its MCP tools can reach.
What we review and deliver
Tool and server trust
Inspect tool discovery, descriptions, update paths, secret handling, and the consequences of trusting a compromised integration.
Execution and data exposure
Map filesystem, network, cloud, and business-system access. Review approval boundaries and the evidence produced by tool calls.
Starting scope
One MCP deployment, up to five tool operations, one authentication flow, and an agreed client integration. Additional servers and test scenarios are scoped separately.
Availability, access, integrations, and complexity are confirmed before engagement.
What you receive
- Documented decisions and priorities
- Actionable recommendations with owners
- Technical and executive summary
- English or Spanish handover session
Before we begin
We agree the authorized systems, test scenarios, deliverables, and acceptance criteria in writing. An architecture assessment is not an exhaustive penetration test. Production testing, implementation, and continuous monitoring are included only when explicitly scoped.
Read the assessment method →
NEXT STEP
Request a scoping call ↗
Define the scope before you commit.
Tell us about your environment, risk concerns, and target date.