Early-stage · Human-led MCP permission audits
Your agents have access. Know how much.
ScopeSentra reviews the MCP servers your AI agents connect to, the tools they expose, and the scopes they've been granted. You get a clear inventory, evidence-backed findings, and a prioritized plan to cut access down to what each workflow actually needs.
Scope, timing, and fees are agreed in writing before any work begins.
- repo:readin use
- repo:writein use
- repo:adminexcess
- tickets:readin use
- tickets:deleteexcess
- warehouse:query:*unclear
- deploy:productionexcess
The problem
Agent access grows quietly. Reviews don't keep up.
MCP makes it easy to give agents real capabilities. That same ease means permissions accumulate faster than anyone tracks them.
Excessive scopes
Servers are often connected with broad tokens to get a pilot working. Those grants rarely get revisited, so agents inherit write, delete, or admin rights that no workflow needs.
Changing tool inventories
MCP servers add, rename, and expand tools between versions. What your team approved last quarter may not match what agents can call today.
Unclear ownership
When a server was set up by one team and is used by three, nobody is sure who approves its access, who reviews changes, or who should revoke it.
How it works
Three steps, done by a person, not a scanner.
Each audit is a focused, human-led review. There is no agent to install and no automated scanning of your systems.
- 1
Map access
Together we list the in-scope MCP servers, the tools each one exposes, the identities and tokens behind them, and the scopes those credentials carry. You share configuration and documentation — never live secrets.
Output: Scoped inventory
- 2
Review permission gaps
We compare what each server is allowed to do with what your agent workflows actually require. Excess scopes, high-impact tools, stale grants, and ownership gaps are recorded with the evidence behind them.
Output: Evidence-backed findings
- 3
Deliver prioritized remediation
Findings are ranked by impact and effort, with specific, practical changes — narrower scopes, tool allow-lists, ownership assignments. We walk your team through them live.
Output: Remediation plan + walkthrough
What you receive
Readable findings, not a wall of alerts.
This is a concept of how audit results could be presented. Today, findings are delivered as a written report and walkthrough. The data below is invented for illustration.
Product concept — sample data
Server inventory
| Server | Tools | Scopes used | Owner | Since review |
|---|---|---|---|---|
| repo-tools | 18 | 6/14 | Platform | +3 tools |
| deploy-runner | 9 | 3/11 | unassigned | — |
| ticket-bridge | 12 | 5/8 | Support Eng | +1 tool |
| warehouse-query | 6 | 2/7 | Data | — |
| docs-search | 4 | 2/2 | Platform | — |
in use granted, unused
Prioritized findings
- HighF-01 · deploy-runner
Production deploy scope held by a token with no named owner
Fix: Assign an owner; move deploy:production behind a separate, approval-gated credential.
- HighF-02 · repo-tools
repo:admin granted; no reviewed workflow calls admin tools
Fix: Reissue token with repo:read + repo:write; allow-list the 6 tools in use.
- MediumF-03 · warehouse-query
Wildcard query scope reaches tables outside the stated use case
Fix: Restrict to the two reporting schemas named in the workflow spec.
- LowF-04 · ticket-bridge
New delete_ticket tool appeared after the last approval
Fix: Disable until reviewed; add tool-list diff to the change checklist.
Server names, counts, owners, and findings above are fictional. They show the level of detail an audit aims for — not a shipping product or results from a real engagement.
Request a Security AuditWhy ScopeSentra
Narrow focus. Clear answers.
Most security tooling tries to cover everything. This audit is deliberately scoped to agent permissions so the output is short, specific, and usable.
Permission-focused analysis
We look at one thing closely: what agents are allowed to do through MCP. Not a general pentest, not a compliance checklist — a deep read of scopes, tools, and the identities behind them.
Readable evidence
Every finding points to the configuration, grant, or tool definition it came from, written so an engineer and a security lead can both act on it without a translator.
Practical remediation
Recommendations are specific and sized to your environment — the exact scope to drop, the tool to allow-list, the owner to assign — ranked so you know what to fix first.
Initial offer
MCP Permission Audit
A focused review of a defined set of MCP servers and the agents that use them. Scope, timing, and fees are agreed in writing before any work begins — no surprise invoices, no open-ended engagement.
Scoped inventory
A documented list of in-scope MCP servers, exposed tools, backing credentials, granted scopes, and owners.
Prioritized findings
Permission gaps ranked by impact and effort, each with the evidence it is based on and a concrete fix.
Live walkthrough
A session with your team to review findings, answer questions, and agree on what to tackle first.
Not included
- Penetration testing or exploit attempts
- Certification, attestation, or compliance sign-off
- Installing software or scanning your systems automatically
- Ongoing monitoring after the engagement
Early-stage, by design. ScopeSentra is a new offering. Early engagements are hands-on, and we'll be upfront about what we can and can't assess for your setup before you commit.
FAQ
Straight answers.
How established is ScopeSentra?
ScopeSentra is early-stage. We are running our first audits now and have no published customers, case studies, or testimonials yet. We would rather tell you that directly than imply otherwise.
What does the audit cover — and what doesn’t it?
The audit covers the MCP servers you place in scope: the tools they expose, the credentials behind them, the scopes those credentials carry, and who owns them. It is not a penetration test, a code review of the servers themselves, or a compliance assessment. We confirm the exact boundaries with you in writing first.
How do you handle credentials and sensitive data?
We never need live secrets. We work from configuration files with secrets redacted, scope listings, tool manifests, and screen-shared walkthroughs. Please do not send API keys, tokens, passwords, or customer data — including through the inquiry form on this site.
Do you install anything or scan our systems?
No. The audit is human-led. There is no agent, scanner, or automated tool connected to your environment.
How much does it cost?
Pricing depends on how many servers and workflows are in scope. After an initial conversation we send a written proposal with scope, timing, and a fixed fee. Nothing is billed until you accept it.
Do you hold security certifications or provide attestations?
ScopeSentra does not currently hold organizational certifications such as SOC 2 or ISO 27001, and the audit does not produce a certification or attestation. Findings are advisory, to help your team make informed access decisions.
Request an audit
Tell us what your agents can reach.
A few details help us judge whether an audit is a good fit. Submitting this form doesn't commit you to anything.
What happens next
- 1.We read your inquiry and reply by email if an audit looks like a fit.
- 2.A short call to understand your MCP setup and agree on what is in scope.
- 3.A written proposal with scope, timing, and fee. Work starts only once you approve it.