The Model Context Protocol (MCP) standardizes how agents call tools and data sources. The MCP Trust Framework (MCPF) adds the missing layer: who those servers are, what they claim to do, and whether they meet your security and governance expectations.
# Quick start (community docs)
# 1) Browse repos and specs
open https://github.com/MCPTrustFramework
# 2) Adopt the trust vocabulary:
# DID identity • VCs • issuers • revocations
# capability snapshots • policy gates
Each MCP server can be identified by a DID, with an issuer you can verify.
Attach attestations: ownership, environment, assurance level, compliance, reviews.
Discover approved servers and revoke or deprecate them centrally when needed.
Enforce “only allow servers meeting X” before an agent ever calls a tool.
Traditional IT did not let unknown endpoints into production without identity, registration, and revocation. MCPF brings the same discipline to AI toolchains.
| Step | What happens |
|---|---|
| 1 | An MCP server publishes a manifest of tools/capabilities. |
| 2 | The server is identified by a DID and can receive VCs from trusted issuers. |
| 3 | A registry lists servers, issuers, and revocations in a queryable way. |
| 4 | Runtimes enforce policy: allow/deny, minimum assurance, environment constraints. |