Concepts¶
The Concepts section explains why MCP Mesh works the way it does — the structural primitives, the runtime model, and the design tradeoffs behind features like DDDI, tag matching, and the registry. It is reference-shaped: you look up a concept when you need to understand a behavior, not when you need to remember a function signature.
Use it alongside the other sections:
- Tutorial walks you through building a real multi-agent application from scratch.
- Reference covers exact API signatures, CLI flags, and environment variables you look up while writing code.
- The Python SDK, Java SDK, and TypeScript SDK sections in the top nav are the language-specific deep references.
Concepts is what you reach for when something works (or doesn't) and you want to understand the mechanism behind it.
The pages are ordered as a discovery arc. Start with the foundation — Architecture and DDDI — then the capability model that resolution is built on (tag and schema matching, service views). From there the control plane (the registry, health and discovery) shows how the mesh wires itself together, followed by what agents do with it — routes, LLM agents, streaming, jobs, and multimodal. The arc closes with state (stateful agents and the in-process escape hatch) and operations (the audit trail). You can read any page on its own, but front-to-back it builds from primitives to surfaces to operations.
Sections¶
-
Architecture core
The framework's structural primitives — how agents, the registry, and dependency injection compose into a mesh.
-
Discovery & matching
How the mesh wires agents together at runtime — heartbeats, capability resolution, and the matching rules that pick a provider.
-
Runtime surfaces
How data flows once agents are wired — streaming responses, long-running jobs with event injection, and the audit trail that records every call.
-
Stateful patterns
Agents that hold state across calls — the session model and the in-process escape hatch for code that can't be made stateless.
See also¶
- Tutorial — guided build of a multi-agent application.
- Reference — API signatures, CLI flags, env vars.
- Python SDK — decorators, dependency injection, LLM integration, FastAPI wiring.
- Java SDK — Spring Boot annotations and integration.
- TypeScript SDK — mesh functions and Express wiring.