liveBook · Manning

MCP

As explained in Architecting for Autonomy

An open protocol for connecting AI applications, models, or agents with external tools, data sources, and contextual capabilities through defined interfaces.

MCP is a communication mechanism for an agent to discover and invoke tools or access resources through defined interfaces. It standardizes tool discovery and invocation, but does not itself define business purpose, delegated authority, decision context, or evidence obligations. section 6.2.3 section 6.3

Treat MCP as one part of a governed interaction:

  • Use the conversational contract to define why the agent acts, what constraints apply, what authority is delegated, and what evidence is required. section 6.2.3
  • Expose bounded enterprise capabilities through an MCP server or gateway rather than granting direct access to underlying systems. section 6.3
  • Enforce identity, scopes, business-purpose policy, validation, recovery, and evidence recording at invocation time. section 6.3

Quality investigation

An agent investigates whether manufacturing lot 2418 requires containment. MCP can expose the production-history tool, while the conversational contract restricts evidence sources and prohibits placing the lot on hold or contacting external parties. A valid tool call therefore remains insufficient if it exceeds delegated authority. section 6.2.3

Table 6.3 API, event, and conversational contracts
Contract type
Primary design concern
Typical contents
What it does not carry by itself
API contract
A callable operation’s request and response shape
Endpoint, method, request and response schemas, status codes, authentication scheme
Runtime intent, delegated authority, decision context, and evidence obligations
Event contract
A published fact, state change, or signal
Topic, event type, payload schema, producer, timestamp, version
Why downstream action should occur, who owns the outcome, and how causation is reconstructed
Conversational contract
The meaning and boundaries of an agentic interaction
Intent, context, constraints, authority, outcome, evidence obligations
Transport mechanism, endpoint implementation, and full payload schema

Compare API, event, and conversational contracts to see which interaction concerns MCP supports and which remain outside its scope. section 6.2.3

How Architecting for Autonomy covers it

How deep it goes
Explains MCP’s host-client-server interaction and governance boundary, but defers full business authority to conversational contracts and wrappers.
Examples
Uses a manufacturing quality investigation involving production history, non-conformance signals, and lot 2418 containment.
What it assumes
Assumes familiarity with agents, APIs, authorization, and enterprise integration while defining MCP’s role from its interaction boundaries.
What you can do after
Operate/configure MCP through bounded tools while separating technical invocation from business authority and evidence obligations.
How it connects
Connects API contracts, event contracts, conversational contracts, A2A, governed tool wrappers, OAuth 2.0, and policy enforcement.

Also explained in

Compare how 21 books explain MCP →

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Architecting for Autonomy. The section links point to the parts of the book it draws on.