MCP Tool Registry
A governed catalog of every tool an AI agent can call — your APIs, your databases, your internal systems — with typed schemas, permission scopes, audit trails, and the standard protocol (MCP) that turns 'we exposed it to the LLM' into 'we know exactly who called what when'.
Production AI is not a prompt. It is a system of context, tools, permissions, traces, evals, and feedback loops.
What it is — and what MCP is
The Model Context Protocol is Anthropic's open standard, released in late 2024 and broadly adopted across the agent ecosystem, for exposing tools, resources, and prompts to AI models over a uniform interface. An MCP server describes its tools with JSON Schema; an MCP client (an agent, an IDE, a workflow runtime) discovers and invokes them. The registry sits between the server and the agent, governing which tools are visible, who can call them, and what is audited.
- JSON Schema input and output contracts
- Permission scopes per tool, per agent, per tenant
- Approval requirements for high-impact tools
- Audit log on every invocation with inputs and result
Why a registry, not just a tool list
Exposing tools is easy. Governing them is the hard part. The registry handles secret boundaries (tools never see raw credentials, only scoped tokens), tenant isolation, deprecation and version pinning, and the mapping between a tool a model can see and the underlying API or database call. When an MCP server changes its schema, the registry flags downstream workflows for re-validation instead of letting a silent break ship.
What it works with
Sits on top of the AI Platform gateway and underneath every Agent Workflow. Tools authored by your team, vendor MCP servers (GitHub, Linear, Slack, internal data platforms), and tools generated by the Self-Optimizing Agents loop all register the same way and are subject to the same governance. Source Graph access boundaries flow through the registry so a tool cannot read data the requesting principal cannot.
When you need it
Signals: AI applications calling business APIs with shared credentials and no audit; an internal copilot whose 'what can it actually do' is debated rather than documented; vendor AI tools quietly accumulating with no central inventory; an auditor asking 'who called what action on whose behalf and when' and your answer involves reading logs by hand.
Related resources
The execution engine that turns an AI agent from a chat-window demo into a long-running, event-driven, restartable process you can trust with real operations.
The policy layer for what an AI system is allowed to read, call, decide, and ship — encoded as configuration the runtime enforces, not as a document on a shared drive.
Trace-level visibility into every model call, retrieval, tool invocation, decision, approval, and failure inside an AI workflow — the substrate every other discipline (evals, optimization, governance) reads from.
A capability in the Group e-media information AI stack. This resource connects the subject to data substrate, agent runtime, evals, and operations.
A capability in the Group e-media information AI stack. This resource connects the subject to data substrate, agent runtime, evals, and operations.