Software catalog MCP server for AI coding agents

An agent that writes code without knowing how your systems fit together guesses at the rest. Componecat gives it the map: the same catalog your team uses, served over the Model Context Protocol, scoped to exactly what the agent is allowed to read.

What your agent can look up

Once connected, an agent has tools to:

• Search and browse the catalog. Find entities, list them by kind or owner, and read one entity in full, with its fields, owner and lifecycle.
• Start from a description. On Pro and Enterprise, with an embedding model configured, an agent can ask for "the service that sends invoices" and get the best matches, each with a snippet saying why it matched.
• Follow relationships. Walk dependencies upstream or downstream, find how two entities are connected, and run impact analysis to see the blast radius of a change.
• Read documentation and interfaces. Pull an entity's documentation, its activity log, and its API specs and package coordinates, to generate a correct client or answer a question with your context.
• Check scorecards. Read an entity's score against your standards and the rules it is failing, with a remediation link for each.
• Follow your wiki. Read the guides you wrote for a kind, team or entity, so your conventions are applied when the agent creates or edits something.

Each entity is also readable as an MCP resource, in the same componecat.yaml format your teams commit.

Connect your client in a minute

The server is listed in the MCP registry as com.shaftware/componecat.

• VS Code and Cursor: one-click install from the docs.

• Claude Code: one command.

  claude mcp add --transport http componecat https://app.componecat.ai/api/mcp
  

• Claude.ai and Claude Desktop: add a custom connector with the endpoint URL.

• ChatGPT: add it in developer mode with OAuth as the connection method.

• Anything else: any client that supports remote MCP servers takes the URL directly.

Step-by-step setup for each client is in Connect AI agents.

Access you grant, not access you hope for

Catalogs are private, so the endpoint requires authentication:

• OAuth with dynamic client registration. Most clients need no manual setup. You approve the request in the browser, and the approval screen names the client that is asking.
• Scoped grants. Give an agent read or write access per resource (catalog entities, teams, schema, wiki), and narrow a token to specific entities when that is all it needs.
• Tools follow the grant. A read-only agent is never offered the editing tools, so it cannot be talked into using them.
• Personal access tokens for scripts and other non-interactive clients.

Context that stays current

An agent is only as good as the catalog it reads. Teams commit a componecat.yaml descriptor next to their code, and repository scanning picks descriptors up across your GitHub, GitLab or Bitbucket organizations on a schedule and on every push. Team sync mirrors owners from GitHub, GitLab, Atlassian or Microsoft 365, so "who owns this?" has a real answer. Integrations come with Pro and Enterprise.

What teams use it for

• Grounded code generation. Before touching a service, the agent looks up its kind, languages, interfaces and conventions.
• Cross-service integration. The agent follows depends-on edges, reads the target's interface spec and writes a correct client.
• Ownership resolution. The agent finds the owning team to route a review or an escalation.
• Standards triage. The agent checks a service's scorecard results and works through the failing rules first.

The MCP server is included on every plan, starting with Free. Compare the plans on pricing.