Agents now act.
They call tools, write to systems, and move data. Their credentials can reach further than the task requires.
AI AGENT & MCP SECURITY · U.S. ARMY VETERANKnow what your
Know what your
AI agents can do
before they do it.
I assess agent and MCP deployments for permission risk, data exposure, and missing oversight, then help you fix what matters first.
Human-led. Evidence-based. Fixed scope.
THE AGENT BOUNDARYILLUSTRATIVE MAP
Data sourceRETRIEVED CONTENT
AI agent
Approval gateCHECK BEFORE ACTION
Connected toolWRITE-CAPABLE
SCOPED IDENTITY · REVIEWABLE EVIDENCE
Can it write more than it needs?
Map the effective identity and each tool’s allowed operations, then test the boundary.
HUMAN-LED REVIEWSCOPE → TEST → EVIDENCE
PERMISSIONS / DATA / OVERSIGHT / EVIDENCE / RESPONSEEXPLORE THE PRACTICE
01 / THE BOUNDARIES HAVE CHANGED
New tools. Real consequences.
The question is what the
whole workflow can do.
MCP connects the risk.
Every connected server adds tools, scopes, and data paths. Those boundaries need a deliberate review.
A scan is one view.
A clean infrastructure scan doesn’t establish that an agent’s actions are authorized or that a person approved the change.
02 / ONE CONNECTED PRACTICEFrom code to runtime.
From code to runtime.
Evidence at every boundary.
Assessment methods.
Engineering you can inspect.
- 01
See
Map agents, servers, tools, identities, and data paths.
Discovery method - 02
Evaluate
Inspect behavior and traces across AI pipelines.
TraceLogicAI - 03
Prioritize
Connect sensitive data, write access, and missing oversight.
Scorecard method - 04
Harden
Narrow identity and check the delivery path.
Aegis + AuditTrail SDK Temporary identity evidence - 05
Respond
Connect detection, controlled response, and evidence.
Project Sentinel
03 / ENGINEERING EVIDENCE
Explore GitHub Built to work. Open to inspection.
Projects and simulations behind the practice. Each record includes its evidence and limits.
EvaluateLive · Active development
TraceLogicAI
Making AI behavior visible. Evaluation, trace evidence, and runtime assurance across AI pipelines.
OUTCOMECompare pipeline behavior, inspect execution evidence, and identify where AI workflows need stronger controls.
RespondCompleted · Maintained
Project Sentinel
A connected security response system. Detect events, orchestrate remediation, and preserve the evidence.
OUTCOMEConnect cloud detection and controlled response with infrastructure governance and operational visibility.
HardenCompleted
Operation Aegis
Security belongs in the delivery path. A Docker-based pipeline for a simulated fintech platform.
OUTCOMEMake security validation a repeatable part of changes moving from pull request to staging.
See · HardenCompleted
AuditTrail SDK
Cloud compliance without static keys. Resource inventory and structured AWS audit evidence.
OUTCOMEInspect AWS resources with temporary credentials and preserve structured evidence for review.
Prioritize · HardenActive development
RevenuePilotAI
A coordinated sales workflow for B2B teams. Specialist AI agents research accounts, ground answers, and prepare CRM actions for human approval.
OUTCOMEConnect account research, opportunity strategy, and follow-up in a sales workflow with human review before CRM changes.
WORKS WITH YOUR STACKFamiliar tools.
Familiar tools.
Explicit boundaries.
Technologies in my engineering practice.
- AWS
- Azure
- Terraform
- GitHub Actions
- Docker
- MCP
- Claude
- Python
- TypeScript
04 / CLEAR SCOPE. USEFUL OUTCOMES.
Start with the right engagement.
MCP Readiness &
Security Assessment
Understand what your agents can reach, what they can change, and where oversight needs to hold.
- Five-dimension readiness scorecard
- Prioritized findings & owner-assigned fix plan
- Reviewable evidence pack
Inquiry Workflow Pilot
Give requests a clear owner. Pilot one bounded workflow for intake, routing, and follow-up.
AuditBuildImprove
Map the handoff gaps, test one workflow with approval boundaries, then review results against an agreed baseline.
Explore the workflow pilot05 / SPECIFICITY IS THE PROOFSee the method.
See the method.
Inspect the evidence.
A U.S. Army veteran and systems builder. My public engineering work connects evaluation, scoped identity, delivery checks, and controlled response.
The sample demonstrates the deliverable. Project records show engineering evidence and limitations.
READINESS SCORECARD
ILLUSTRATIVEControls you can explain.
Permissions1 / 3 Defined
Data Access1 / 3 Defined
Oversight1 / 3 Defined
Evidence2 / 3 Verified
Response0 / 3 Absent
Fictional example, not a client result. Unknown evidence is marked separately. No aggregate pass/fail score.
06 / THE PERSON BEHIND THE SYSTEMSSecure systems.
Secure systems.
Human impact.
I’m Malik Dixon. I connect AI, cloud, and security to build systems that work for people. The work starts with the mission, the workflow, and the decisions a person needs to make.
As a U.S. Army veteran, I bring a systems mindset: understand the mission, build with intention, and verify the result.
Open to opportunities · Remote / Worldwide
Discuss an engineering opportunity HOW I THINK / HOW I BUILD
01
Start with the workflow.
Understand the people, friction, and outcome before choosing the tool.
02
Make trust explicit.
Give systems clear permissions, observable behavior, and meaningful oversight.
03
Build for the next person.
Readable code. Useful documentation. Systems that are easier to maintain.
07 / FIELD NOTES
All insights Agent boundaries. Engineering judgment.
FIELD NOTES / 01
Agent & MCP security6 min read
What an MCP Security Assessment Should Actually Prove
Read the noteFIELD NOTES / 02
Agent & MCP security2 min read
Review the permissions behind the tool.
Read the noteFIELD NOTES / 03
Agent & MCP security2 min read
A connection is not a permission review.
Read the note08 / BEFORE WE START
Clear scope. Useful answers.
Do we need this if we already have a CNAPP or cloud scanner?
Possibly. Start by checking what your existing platform already covers. This assessment focuses on a bounded agent workflow: effective tool permissions, sensitive data paths, enforced approval, and execution evidence. It can complement continuous cloud monitoring. If your current team and tools already verify those controls, a separate assessment may add little.
What access do you need, and is it read-only?
Discovery starts with redacted architecture, tool manifests, permission policies, and representative traces. Prefer customer-provided exports, a screen share, or scoped read-only access. Any behavior testing uses an agreed test environment and explicit authorization. Production writes, credentials, and hardening are never assumed to be included.
How long does an assessment take, and what do we get?
The delivery window and price are agreed after a fit conversation, once the number of agents, servers, workflows, and test cases is clear. The fixed scope includes a five-dimension readiness scorecard, prioritized findings, a fix plan with owners and effort estimates, and an evidence pack. Missing evidence and untested areas are identified explicitly.
Can you help fix the findings, or just report them?
Hands-on hardening can be agreed as a separate phase. That may include narrowing tool permissions, binding approval to the exact proposed action, improving trace evidence, and exercising access revocation. Each change has an owner, acceptance checks, and agreed access before implementation.
Who is Malik Dixon?
Malik Dixon is an engineer, systems builder, and U.S. Army veteran working across AI systems, cloud security, DevSecOps, and full-stack development. He offers human-led AI agent and MCP security assessments and a focused inquiry-workflow pilot for small B2B service teams.
MAKE THE BOUNDARIES EXPLICITFind out what your agents
Find out what your agents
can actually do.
A clear scope. A practical fix plan. Evidence you can use.