Know in one week whether your MCP platform is safe to ship.
You moved fast on AI agents. Good. Now production is on the horizon and the question nobody on your team can answer with confidence is: "Is this actually safe?"
Not a PDF. A call. One clean answer you can act on.
The guarantee
3 production-relevant findings your team hadn't already flagged — or you pay nothing.
Every team that moves fast on MCP has structural gaps — not because they're sloppy, but because the tooling doesn't enforce governance by default. I've never completed one of these reviews and found fewer than 3 things. The guarantee is there because I'm confident in that.
"It works in dev" ≠ "it's safe in production"
The gaps most teams hit — not because they're careless.
The tooling doesn't enforce governance by default. These are structural gaps, not team failures.
-
Ungoverned tool access
Agents call functions, read data, and hit systems with no per-function rights model. One over-permissioned tool is one prompt-injection away from an incident.
-
No fail-close posture
When your gateway, a model, or an MCP server degrades, does the system fail closed (refuses safely) or fail open (silently does the wrong thing)? Most teams don't know.
-
Ad-hoc onboarding
Adding a new MCP server is a Slack message and a config edit. No review gate, no audit trail, no declarative source of truth.
-
No observability
When something misbehaves, you can't answer: what did it call, with whose rights, and why?
-
Secrets and identity sprawl
Credentials embedded in configs, no clean IDP tie, unclear who-can-do-what.
You can put an AI agent in front of your users. The question is whether you can put it in front of your security team and your board.
The offer
MCP Production-Readiness Audit — $10,000 flat, 1 week.
A structured review across the 7 dimensions that determine whether an MCP/agent platform is production-grade. At the end of the week, you have a named verdict: go, or not yet — and if not yet, exactly what to fix first.
Seven dimensions assessed
-
Tool-access governance / RBAC
Per-function rights, least-privilege posture, blast radius of each tool. One over-permissioned tool is one prompt-injection away from an incident.
-
Fail-close vs fail-open
Degradation behavior of your gateway, models, and MCP servers. Does the system refuse safely or silently do the wrong thing?
-
Onboarding flow
How new MCP servers and tools enter production — declarative vs ad-hoc, review gates, audit trail.
-
Observability and tracing
End-to-end visibility across agent → gateway → model → tool. Can you reconstruct any call after the fact?
-
Multi-LLM routing and cost controls
Virtual-model routing policy, fallback behavior, per-team cost attribution, latency guardrails.
-
Security and secrets
Secret handling, IDP integration, identity propagation, prompt-injection and exfiltration exposure.
-
Production table-stakes
Rollout readiness, kill-switch, rate limits, eval gates — the operational requirements for shipping with confidence.
Five deliverables
- ✓
Readiness Report
Findings across all 7 dimensions, evidence-backed. Written to be useful for your engineers and readable by your security team and leadership.
- ✓
Scored Gap Matrix
Every dimension rated red / yellow / green with severity. The state of your platform on one screen — defensible and re-runnable.
- ✓
Go / No-Go Verdict
A named, explicit call. Either "you're ready to ship with these caveats" or "not yet — fix these three things first." The kind of answer you can act on.
- ✓
Prioritized 90-Day Roadmap
Sequenced remediation ordered by risk-reduction-per-dollar. What to fix this week, this month, this quarter. Effort-tagged so you can staff it.
- ✓
Live Review Session
A 60-minute working session with your team to walk the findings, pressure-test the roadmap, and align on the highest-leverage next step. Not a presentation — a working session.
What you provide
- Read-only access to relevant configs and repos
- A technical point of contact for the week
- ~3 hours of engineer time for interviews + the review session
Payment
3 production-relevant findings your team hadn't already flagged — or you pay nothing.
Every team that moves fast on MCP has structural gaps — not because they're sloppy, but because the tooling doesn't enforce governance by default. I've never completed one of these reviews and found fewer than 3 things the team hadn't already documented. The guarantee is there because I'm confident in that, and because you shouldn't have to take my word for it.
Pricing
$10,000 flat. One week. No meter.
$10,000
Payment milestones
Proceed to implementation within 60 days? The full $10K credits toward the implementation engagement. You pay for the diagnosis once. If you build, the diagnosis was free.
For context
A traditional MCP or LLM security pentest at a specialized firm runs $30–60K over 4 weeks.
It produces a PDF report. No go/no-go verdict, no implementation roadmap, no live session with your team.
This audit delivers a named verdict, a prioritized roadmap, and a live review in one week at $10K.
If you proceed to implementation within 60 days, the full $10K credits toward that engagement — you pay for the diagnosis once.
Not ready to commit to the full audit?
A no-commitment option at every level.
Start where makes sense. Stop whenever the next level isn't needed yet.
Run mcp-gateway-scan in your repo. No sign-up. See your surface-level gaps across the 7 dimensions in minutes.
npx mcp-gateway-scan30-minute focused session. I look at your scanner output and name your top 2–3 risks. Real work, not a sales call.
Email to scope →One week. Named verdict. 90-day roadmap. The full offer described on this page.
Book a scoping call →"Production-ready or we keep working until it is." Scoped from the audit roadmap; the $10K fee credits toward this.
Email to scope →Re-audit on the same methodology each quarter plus drift alerts wired into your CI. For teams that want ongoing readiness, not a one-time check.
Email to scope →Why Willian Pinho
I build and ship the protocol this audit is about.
-
Maintainer of the open-source MCP scanner
I maintain
mcp-gateway-scan— the CLI you may have just run — andlarge-file-mcp, an MCP server that works across Claude Code, Cursor, and Gemini CLI. I build the protocol this audit is about. -
Just did exactly this work
Recently operationalized an MCP-services gateway for a Security Engineering team: YAML-in-GitHub MCP onboarding, per-function rights management, multi-LLM routing, OpenTelemetry observability, Okta IDP integration. This audit is that engagement made repeatable.
-
16 years in production engineering
Including a payment SDK serving 10M+ users at 99.9% uptime. I've been the person who decides whether an AI capability is safe to ship — and built the evaluation frameworks behind that call.
-
Security-first background
Extensive enterprise background including Security Engineering leadership. I know what security teams actually ask — and what "production-ready" means to a board-level risk conversation.
Frequently asked
Questions
How does the week work?
Day 1: kickoff, access provisioned, interviews scheduled. Days 2–4: review across all 7 dimensions, questions to your team as they come up. Day 5: report and gap matrix delivered. Days 6–7: live review session scheduled at your convenience.
Read-only — what does that mean exactly?
I review configs, repos, architecture diagrams, traces, and logs you share. I never touch production systems directly. Fault tests (if any) run in staging with your written sign-off in advance. No access to secret values — references only.
What if we're not in production yet?
This is also the right time. It's faster and cheaper to fix the gaps before you ship than after. The audit produces a launch checklist, not just a retrospective.
What if we're a small team with a simple setup?
The scope adapts to what you actually have. If you're running two MCP servers and a single LLM, the assessment is shorter and the price reflects that. Email me with your setup and I'll tell you whether the full audit is the right fit or whether a smaller engagement makes more sense.
What do we do with the roadmap after?
It's yours. Your team executes it, or we scope a fixed-price implementation engagement. Either way, the roadmap is built to be handed off — sequenced, effort-tagged, no dependencies on me to read it.
Is this a pentest?
No. A pentest finds exploitable vulnerabilities through active probing. This audit assesses architectural posture — governance model, observability, fail-close behavior, secrets handling, onboarding process. Both matter. They're different engagements. If you need a pentest too, I can tell you what to hand to the pentester and what questions to ask them.
If you proceed to implementation within 60 days, what happens to the audit fee?
The full $10K credits toward the implementation engagement. You pay for the diagnosis once. If you build, the diagnosis was free.
Ready to schedule a kickoff?
I run a small number of these each quarter. Email or book a call — I'll tell you if there's a slot available and whether the audit is the right fit for where you are.
No deck, no obligation. Fixed scope, fixed price, written scope statement before any audit work begins.