Model Context Protocol — MCP
An open protocol that gives AI models a uniform way to connect to external tools, data sources, and services. Anthropic released it in November 2024 and donated it to the AAIF (Linux Foundation) in December 2025.
Definition
MCP standardizes the wire between client (the model) and server (the tool) on JSON-RPC. The server exposes tools, resources, and prompts as capabilities; the client calls them as needed.
The authoritative specification at the time of writing is 2025-11-25. In 2026, MCP Apps (SEP-1865) extended the standard to deliver interactive UIs — React-based dashboards, forms, visualizations — from MCP servers to host apps like Claude and ChatGPT, on top of the existing text/structured-data wire.
Governance moved into the Agentic AI Foundation (AAIF), co-founded by Anthropic, Block, and OpenAI under the Linux Foundation — making MCP vendor-neutral. Implementations rolled out concurrently in Claude, ChatGPT, and major IDEs.
Lemma Oracle implementation
Lemma itself runs an MCP server, exposing ZK proof generation, provenance verification, and selective disclosure as callable MCP tools to any AI agent.
Every MCP tool call lands in an audit trail: which model invoked which tool with which authority, in what order. Combined with A2A, the full agent-collaboration surface becomes auditable.
Embedding MCP into a verifiable AI pipeline makes the boundaries between model, tool, and data cryptographically verifiable end to end.