Tool Inventory Matrix
Inventory of MCP servers, tools, permissions, OAuth scopes, credentials, data touched, and allowed actions.
MCP & Tool-Use Blast-Radius Review
Independent MCP security review for teams connecting LLMs to real tools.
A technical review of Model Context Protocol servers, clients, tools, tokens, and approval paths before broad rollout.
Buyer problem
Your LLM clients are gaining access to real tools, but the team cannot yet show which actions are permitted, which trust boundaries are crossed, or how destructive actions are contained.
Best fit: Platform, security, developer experience, and AI infrastructure teams adopting MCP or custom tool-use gateways.
Scope
MCP clients and servers, tool schemas, OAuth and delegated access, credentials, data flows, read/write/delete/send/execute permissions, approval gates, logging, and rate limits.
Deliverable
A concrete work product tied to the decision.
A tool inventory and trust-boundary review with a permission matrix, destructive-action control plan, blast-radius map, and prioritized scope-reduction roadmap.
Timeline
Typically 10 business days from kickoff and access confirmation.
Signature view / permission lattice
Every tool call should declare its action class.
MCP risk becomes easier to discuss when every tool is classified by what it can do and paired with the control required before that action is allowed.
ActionAuthority questionReview focus
READ
Can this identity see it?
Retrieve records, files, messages, or system state.
WRITE
Can this identity change it?
Create or modify records, files, configurations, or tickets.
DELETE
Can this action be recovered?
Remove data or resources where recovery may be limited.
SEND
Can this action leave the system?
Communicate externally through email, messaging, publishing, or transactions.
EXECUTE
Can this action alter operations?
Run commands, deploy changes, approve workflows, or alter operations.
Sample assessment package
Agree the review boundary, authorized methods, and decision the work must support before starting. The scoped package draws from the outputs below and identifies untested paths and unresolved assumptions alongside the findings.
MCP Trust-Boundary Review
Analysis of MCP client/server trust, upstream APIs, token flow, session handling, and local-server exposure.
Destructive-Action Control Plan
Classification of high-impact actions and required approval, policy, sandboxing, or hard-deny controls.
Blast-Radius Map
Clear explanation of what an abused agent/tool path can read, write, modify, delete, send, execute, or deploy.
Scope Reduction Plan
Practical roadmap for least privilege, allowlists, user-scoped auth, egress limits, and tool-call logging.
Risk addressed
Tool, data, and trust-boundary risk in one review.
- Focused on real tool execution paths rather than generic LLM policy language
- Suitable for internal platforms, customer-facing integrations, and vendor review preparation
- Balances permission reduction with the workflow utility teams need to preserve
When to use it
Technical review before a consequential decision.
Use this when LLM clients are being connected to internal systems, SaaS tools, file stores, ticketing systems, databases, or deployment workflows.
Next step