Let your agent read and update architecture docs

An agent asked to add a service needs to know which database already exists and how the system is documented. ArchDocs exposes documents and catalog objects over MCP so the agent can read that context before it writes a diagram.

Add a refunds service

Consider a fictional booking system. You ask your agent: "Add a refunds service that reads from the bookings database."

The agent can search for the existing booking documents, read their YAML and look up the database in the object catalog. It can reuse that object rather than create a separate database with the same name. Then it validates a proposed diagram and writes it through the MCP server. The server compiles the YAML into canvas elements.

You review the resulting diagram in ArchDocs. Open the database's dependency view to inspect the references recorded across your documents. Those dependencies come from documentation. They do not prove what is running in production.

The request is an example workflow, not a claim that an agent always chooses the right object or produces a correct design. Check its references and relationships before relying on them.

Read, validate and write

The hosted MCP server has tools for documents and objects. The list below follows the server's tool registrations. The tool reference has input fields and behavior for each call.

Documents

• list_docs lists documents available through the document overview.
• search_docs searches documents by name.
• get_doc reads a document as YAML.
• create_doc creates a document from YAML.
• update_doc replaces a document's YAML.
• patch_doc applies changes to named nodes, relationships or fields.
• validate_yaml checks YAML without saving a document.
• get_schema returns the DSL schema and construct reference.

update_doc replaces the existing DSL content. Ask your agent to read the current document first and review the proposed change. patch_doc supports smaller edits, but a valid patch can still describe the wrong architecture.

Objects and placement

• list_object_kinds lists object kinds and their preset document slots.
• create_object creates a catalog object.
• list_objects lists objects, with optional name or space filters.
• get_object reads an object's details and documents.
• list_object_docs lists an object's preset slots and custom documents.
• list_spaces lists spaces available to your account.
• assign_doc_to_object attaches a document to an object.
• assign_doc_to_space files a document in a space or moves it to Drafts.
• assign_object_to_space binds an object to a space or removes that binding.
• release_doc_from_object detaches an object's document back to the document lists.

Placement affects visibility. In particular, binding an object to a public space makes it readable by anyone holding that space's share link. Check the destination before asking an agent to move content.

Connect your client

The MCP connection guide explains browser authorization. Follow the setup for Claude Code, Cursor, Codex or VS Code. Add the hosted server URL in your client and approve the OAuth connection in your browser.

Your client receives the content the tools return. Choose a client whose data handling fits your documents, and review its requests to change content. MCP gives an agent access to the documented system; it does not verify the design for you.

AI inside ArchDocs

You can also generate a diagram from a description or ask questions about your architecture inside ArchDocs. Diagram generation uses AI credits. Check the linked documents and generated diagram against the system you maintain.

For system design interviews, AI can draft kits, suggest probes and summarize evidence. AI never scores candidates. Every rating and hiring recommendation comes from a human.