Draft of the Next MCP Specification (github.com)

The Model Context Protocol (MCP) project has published a draft specification in its public GitHub repository, laying out a structured blueprint for exchanging model context metadata. The draft lives under docs/specification/draft and includes dedicated sections for architecture, core/basic concepts, client and server behaviour, plus a schema file and changelog — signalling a spec designed for both implementers and integrators. The repo also maintains versioning and dated releases elsewhere, indicating an intent for iterative, versioned evolution. This matters for the AI/ML community because a common, formalized protocol for model context (inputs, metadata, provenance, runtime hints) lowers integration friction across frameworks, inference services, observability tools and governance systems. By separating architecture, client and server responsibilities and publishing a schema, the draft makes it straightforward to build SDKs, middleware, and interoperable services that can share contextual information reliably — improving reproducibility, auditing and tooling interoperability across deployments. Key technical implications: implementers can follow defined client/server contracts, validate payloads against the published schema, and track breaking changes via the changelog/versioning approach. The draft’s structure promotes spec-driven implementations and community feedback before stable release, positioning MCP as a potential standard for context propagation in production ML stacks.