MCP (Model Context Protocol)
The protocol through which an agent reaches external tools, servers, and data beyond its own reasoning. GovTech's phase one excluded MCP access entirely, restricting the agent to built-in reasoning inside its own sandboxed environment; phase two reintroduces it through a governed, sandboxed mechanism rather than an open connection.
Defined in 3 GAGE programs, which carry 10 distinct definitions of it. The wording above is taught in Certified Agentic AI Governance Professional (CAAGP).
How each discipline defines it
The same term does different work depending on who is using it. These are the definitions as each program teaches them, unedited.
A named protocol, cited in the MGF's agentic value chain as an example of a tooling-provider offering, that lets an agent connect to external tools and data sources in a standardized way. Named here as a value-chain example; its own protocol-layer threats are developed in full in this program's control-and-threat module (see Topic 5.6).
An emerging open standard, still maturing as of August 2026, for connecting AI agents to external tools and data sources in a structured, interoperable way; several identity platforms, including Auth0, now treat MCP servers as first-class identities subject to their own authorization and logging controls.
A common standard for connecting AI agents to external tools and data. Notable here because its open-by-default tooling can over-expose access unless scoped, the condition the Invariant Labs GitHub incident exploited.
The named standard for agent-to-tool communication, introduced by Anthropic in November 2024 to let an AI agent connect to external tools and data sources without custom integration code for each one; adoption has moved fast, which is also why a flaw in its trust model can affect many implementations at once.
A named communication protocol for connecting an AI agent to external tools and data sources, referenced by the WEF paper's protocol-layer treatment; full technical treatment is out of scope for this topic. (see Topic 1.2)
Where it is taught
The exact lessons this term appears in. The first module of every program is free with a free account.
- Securing Agent Permissions and Access Boundaries · AI Security Fundamentals, Certified AI Practitioner: Workplace Foundations
- The Parts That Make an Agent Act · The Agent, Deconstructed, Certified Agentic AI Governance Professional (CAAGP)
- Why Singapore's Framework in Ohio · The Agent, Deconstructed, Certified Agentic AI Governance Professional (CAAGP)
- The Whole Framework, Worked · The Agent, Deconstructed, Certified Agentic AI Governance Professional (CAAGP)
- Action-Space · Capability and Autonomy, Certified Agentic AI Governance Professional (CAAGP)
- Agent Identity · Bounding by Design, Certified Agentic AI Governance Professional (CAAGP)
- The Agentic Value Chain · Meaningful Human Accountability, Certified Agentic AI Governance Professional (CAAGP)
- Protocol-Layer Threats · Technical Controls and Threat Modeling, Certified Agentic AI Governance Professional (CAAGP)
- Ship the Agent in Stages, Never All at Once · Rollout, Monitoring, and Incident Response, Certified Agentic AI Governance Professional (CAAGP)
- The agent's trail: who acted, on whose authority, using what, and why, recorded for every autonomous touch of your estate · Access and the Keys, Certified AI Data Governance Professional (CADGP)
Where it shows up at work
A definition is where MCP (Model Context Protocol) starts. The lessons above teach it inside the competencies below, which are what job descriptions ask for and what the GAGE credential records. Every link here is drawn from the same careers map that builds a role's readiness check.
Governs AI agents that take actions: tool access, least privilege, interruptibility, cascading actions and accountability for what an agent did.
3 lessons on this page teach it as a primary outcome.
Understands prompt injection, data poisoning, model theft, insecure integrations and excessive agent privileges, and the controls that reduce each.
2 lessons on this page teach it as a primary outcome.
Applies IMDA's Model AI Governance Framework, AI Verify and the PDPA alongside Western frameworks for organizations operating across Asia.
One lesson on this page teach it as a primary outcome.
Roles that hire for it
Each role page shows the full competency set, what the seat decides, and a readiness check against your own evidence.
- Chief Information Security Officer, AI security focusneeds it at depth
- AI Security Architectneeds it at depth · 8 open postings this week
- Chief Technology Officer, AI engineering editionneeds it at depth
- AI Governance Engineerneeds it at depth · 7 open postings this week
- Agentic AI Engineer, orchestration editionneeds working knowledge
- AI Evaluation Specialistneeds working knowledge · 6 open postings this week
How job descriptions say it
Postings rarely use the textbook word. These are the phrases the careers map reads as asking for the competencies above.
Where the frameworks place it: IMDA Model AI Governance Framework for Agentic AI; NIST AI RMF; OWASP Top 10 for LLM Applications; Google SAIF.
One interview question about it
An agent can send emails and update records. What may it touch, what needs a human, and how do you prove afterwards what it did?
Asked of candidates for agentic ai controls and authorization boundaries. Answering it in your own words is exactly how a GAGE topic is passed, and the credential records the pass where an employer can check it.
If you come from another field
Some backgrounds already carry this competency, and the readiness check gives credit for it.
- Cybersecurity and IT carries it as a claim to back up: Threat modeling and control design transfer; the AI-specific threats are new.
- A computer science or software engineering degree carries it as a claim to back up: Security fundamentals are a course you took, not yet a control you ran.
- A cybersecurity or information assurance degree carries it as work you can show: Security fundamentals are your coursework and your labs.
Terms it appears with
Not an alphabetical neighbourhood: these are the terms taught in the same lessons, ranked by how often they appear together.