DEVELOPERS

Build on engineering reality

Today (v1.0): run the local API with contextmint serve and integrate via REST. Roadmap: MCP tools and a fuller CLI for blast radius, packs, and CI gates.

API Reference

FastAPI backend — health, chat, search, packs, quality, and actions. OpenAPI at /docs when the server runs.

v1.0 · local server

CLI

contextmint serve starts the API today. Index, health, and graph commands are planned for CI and platform automation.

serve available more commands roadmap

MCP Server

Model Context Protocol tools for blast radius, features, and decisions — agent-native read contracts on the engineering graph.

roadmap · verification engine

Quick start (v1.0)

      # Install extension + Engine — see getting started

      # Or from a dev checkout:

      contextmint serve

      # Verify health

      curl http://localhost:8000/api/health

Full setup: Getting started · User docs: Documentation · Terminology: Glossary

Terms explained (plain language)

ContextMint uses precise engineering words. Here is what they mean for your team — without assuming graph-theory background. Full v1.0 terms: User glossary.

Blast radius: If you change or break X, what else is affected? ContextMint walks real dependency links (imports, service calls, shared modules) to list downstream files, services, and features — like a ripple map, not a guess from an LLM.
Verification Engine: The future product direction after v1.0: automatically check every PR or AI-generated change against ownership, impact, architecture rules, and policy before merge. v1.0 lays the foundation (quality map, gates, local chat); PR-wide verification is on the roadmap.
Engineering Reality Graph (Fabric): A living map of your repo and org facts: who owns which files, how modules depend on each other, recorded architecture decisions (ADRs), and policy rules. Chat and agents can cite this map instead of inventing answers.
Repository DNA: A one-page health summary of a codebase — structure, owners, hotspots, and conventions — so a new engineer or reviewer understands the repo in minutes instead of weeks.
Decision graph: Links why the code looks the way it does: architecture decision records (ADRs), approved PRs, and issues connected to the files they affect. Answers “why do we do it this way?” with evidence.
Drift: When live code diverges from agreed architecture or conventions. Bad drift is an unintended violation; accepted drift is an approved change recorded so alerts do not fire again.
MCP (Model Context Protocol): A standard way for AI tools (Cursor, Claude Desktop, custom agents) to call read-only functions on your engineering data — like get_blast_radius — with structured JSON instead of scraping chat text. Planned for a future release; not connectable in v1.0.
Agent governance: Rules for AI agents that touch your repo: which tools they may call, what they may read, and that every write still goes through human-reviewed patch preview — no silent auto-apply.
Gate 1 / Gate 2: Gate 1 = Context Lens — you approve evidence before the model runs. Gate 2 = patch preview — you approve diffs before anything writes to disk. Ships in v1.0.
Quality map: Stored scan results (duplicate code, structure signals, patterns) shown in Context Lens and governance UI. Answers come from findings on disk, not model imagination. Ships in v1.0.

Need integration support?

Design partners get help wiring the local API and preview MCP contracts on the roadmap. Intake opens by start of September 2026.

Book a Demo Trust & Security