Workshops: Secure Agentic Coding & Governance
MCP Server Governance: Vetting, Allowlists & Approval Workshop
Developers connect MCP servers to their coding agents in minutes, often by running a package straight from a public registry with their own credentials. Each server extends what an agent can read and do, sometimes far beyond what anyone intended. This half-day workshop helps security, platform and engineering leads set up proportionate MCP governance: a vetting checklist, risk tiers, an internal registry, technically enforced allowlists and ongoing monitoring.
Who is this workshop for?
- Platform and developer experience teams
- Security architects and AppSec teams
- IT governance and tool approval boards
- Engineering leads whose teams use Claude Code, Cursor, GitHub Copilot or similar agents
- Owners of internal MCP servers
What you will take away
- Assess a third-party MCP server against a structured checklist covering publisher, code, permissions, data flows and update behaviour.
- Classify MCP servers into risk tiers and attach matching approval requirements to each tier.
- Design an internal registry in which approved servers are easier to find than unapproved ones.
- Enforce allowlists through the managed configuration your agents and IDEs support, and pin approved versions.
- Monitor MCP usage, detect changed tool definitions and revoke a server quickly when needed.
Agenda
Module 1
The MCP risk picture
- Local and remote servers, transports and where the code actually runs
- Tool poisoning, tool definitions changed after approval, prompt injection via tool output
- Overly broad OAuth scopes, shared tokens and data leaving the company
Module 2
Vetting and risk tiers
- Exercise: reviewing two real third-party servers with the checklist
- Tiers: read-only internal, write access, external data, production systems
- Who approves what: security, data protection, system owner
Module 3
Registry and enforcement
- An internal catalogue with owners, versions and usage notes
- Allowlists and managed settings where clients such as Claude Code or GitHub Copilot support them
- Containers, pinned versions and gateways for central authentication and logging
Module 4
Operations and re-certification
- Logging tool calls, alerting on new or changed tools
- Periodic review, deprecation and emergency revocation
- Exercise: drafting your MCP approval process on one page
Why MCP needs its own governance
Traditional software approval looks at an application once. MCP servers work differently: their tool descriptions go straight into the model's context, their output can steer the agent, and a server can change its tools after approval. A package that looks like a harmless documentation lookup can still read local files or send data elsewhere.
A ban is rarely the answer, because the productivity gain is real and teams find workarounds. Effective governance makes the approved path the most convenient one.
What makes a good approval process
We build on what you already have, such as a software approval board or ISO 27001 supplier processes. neooutput AI Solutions Architects lead the session, remotely or on-site, in English or German.
- Clear risk tiers instead of one heavyweight review for everything
- A fast lane for read-only servers from trusted publishers
- A named owner for every approved server
- Technical enforcement rather than a wiki page
- Re-review when versions or scopes change
Frequently asked questions
Is an MCP server from a well-known vendor automatically safe?
Official servers from established vendors lower the risk but do not remove it. You still decide which tools and scopes to allow, and prompt injection can arrive through the data the server returns. The review can be lighter, but it should not be skipped.
Can we enforce an allowlist technically?
Several clients offer admin-managed configuration or policies for MCP servers; what is possible differs by tool and plan. Where a client offers no enforcement, network rules, gateways and endpoint management can close the gaps. We check what works with your tool set.
How is this different from your MCP server development workshop?
That workshop teaches developers to build servers. This one is for the people who decide which internal and external servers may be used in the company and how that stays under control over time.
Workshop
Get in touch
Do you know which MCP servers your developers' agents are connected to right now?
Tell us roughly how many people will join and pick dates that work – we will get back to you within 24 hours with a proposal for content, format and schedule.
Prefer to talk first?
15 minutes, free and without obligation – with our AI Trainers and AI Code Engineers.
Book a call for free (opens in new tab) +49 761 519 751 50 (24/7)Request a training / workshop
Workshop: MCP Server Governance: Vetting, Allowlists & Approval Workshop · Half day (approx. 4 hours)
More workshops
- Workshops: Secure Agentic Coding & GovernanceWhen a Coding Agent Goes Wrong: Incident Response Workshop
- Workshops: Secure Agentic Coding & GovernanceTechnical Debt in AI-Generated Code: Maintainability Workshop
- Workshops: Secure Agentic Coding & GovernanceSandboxing Coding Agents: Permissions & Least Privilege Workshop
- Workshops: Secure Agentic Coding & GovernanceKeeping Secrets Away from Coding Agents: Half-Day Workshop
- Workshops: Secure Agentic Coding & GovernancePrompt Injection Against Coding Agents: Attacks & Defences
- Deploy, Convert & ScaleFrom Vibe-Coded Prototype to Production: The Complete Checklist
