MCP tools & contracts
Planned capability. Expose deterministic engineering facts to AI agents via the Model Context Protocol (MCP) — — blast radius, features, and decisions as structured tools, not chat guesses.
What is MCP?
MCP is an open protocol so AI clients (Cursor, Claude Desktop, internal agents) can call tools with typed inputs and JSON outputs — like REST for agents. ContextMint's MCP layer will expose read-only graph queries your security team can scope with RBAC. Writes still use human-gated patch preview.
Planned tool catalog
get_blast_radius
Blast radius — list impacted services, files, and teams if a node changes or fails. Maps to /fabric/explain or a pre-computed dependency closure.
“What breaks if we remove Payments API?”
get_feature
Feature registry lookup — entry points, owners, and related graph nodes for a named product feature.
get_decision
Decision graph trace — ADRs and architecture intent with evidence ledger and verification status.
get_trace
CI/CD path — deployment → feature → failure linkage (enterprise, after external-memory graph work).
Response contract
Every tool response will validate against AgentContextContract — a fixed JSON schema so agents and CI gates can trust fields like verification_status and evidence paths. See Agent governance.
Example client config (when shipped)
Illustrative only — endpoint and auth will match your deployment profile (local Engine or sovereign server).
"mcpServers": {
"contextmint": {
"url": "http://localhost:8000/mcp",
"headers": { "Authorization": "Bearer ${CONTEXTMINT_TOKEN}" }
}
}
}
Discuss your AI governance challenges
See the local-first decision-control plane — Context Lens, quality map, and human apply gates. v1.0 ships end of August 2026.