ComponecatEngineering context for AI agents: give them the map
An AI coding agent sees the repository in front of it. Engineering context is everything else: which services exist, who owns them, what depends on what, how to call them, and the conventions your team follows. Componecat keeps that context in one catalog and serves it to your agents over MCP.
What counts as engineering context
Six kinds of fact, each one modeled in the catalog rather than scattered across wikis, READMEs and chat:
- Inventory. Every system, service, library, datastore and tool, as an entity with a kind and the fields that kind defines.
- Ownership. Every entity has an owning team, so "who do I ask?" has an answer an agent can look up.
- Relationships. A parent and child hierarchy plus typed dependencies such
as
depends-on, which impact analysis follows to show what a change affects. - Interfaces. The APIs a service exposes, with protocol, URL and auth, and the packages and CLIs a library publishes, with registry and coordinates.
- Documentation. Overview, architecture, runbook and ADR sections on the entity itself, searchable and readable by agents.
- Conventions and standards. Wiki guides for each kind, team or entity, and scorecards that check entities against your standards.
Why the repository is not enough
The code in front of an agent describes one component. The questions that decide whether its change is right usually point elsewhere: which client library to call, what that service's API expects, who has to review a change, what breaks downstream. Those answers live in the relationships between repositories, which is exactly what a catalog records.
Context that stays current
Context that is out of date is worse than none, so the catalog keeps itself in step with the systems that already know the facts:
- Teams commit a
componecat.yamldescriptor next to their code. Across GitHub, GitLab and Bitbucket, a scheduled scan finds new descriptors and every push re-reads a changed one. - Documentation sync sources an entity's documentation sections from Markdown in the repository.
- Team sync mirrors owners from GitHub, GitLab, Atlassian or Microsoft 365.
Integrations come with Pro and Enterprise. On every plan you can create and edit entities by hand in the app.
How agents read it
The catalog is served over the Model Context Protocol. An agent can search it, follow relationships, read an entity's interfaces and documentation, and check the guides for a kind before it creates or edits anything. On Pro and Enterprise, with an embedding model configured, it can also describe what it is looking for in plain language. The MCP server page covers what an agent can do and how to connect each client.
Inside the app, the AI assistant answers the same questions ("who owns the payments service?", "what depends on the auth gateway?") from the catalog itself, on the model you choose.
Start with the services that matter most
The Free plan holds up to 50 catalog entities for up to 5 users, with the MCP server included. That is enough to model one system and the services around it. Getting started takes you from sign-in to a catalog your agents can read.