Core workflow
How to reason with your LLM
The whole point of ContextGraph::ai is to hand your AI a map it can actually read — so it reasons over your architecture instead of guessing from scattered files. Here is the end-to-end loop.
1
Design the base
Map your domains, entities, and the dependencies between them on the canvas. Even a rough first pass gives the agent real structure to reason from — refine it as you go.
2
Add AI Context
Attach a note to any node, link, or the canvas itself. These notes are binding directives — the agent treats every one as authoritative, so this is where you encode the rules it must follow.
3
Share it
Export ▾ → ⇡ Share publishes a read-only viewer plus a machine-readable AI Build extract, and makes the diagram visible to your MCP clients.
4
Connect the MCP server
One-time setup — see Setting up your MCP. Once connected, your agent can list and read your shared diagrams in a conversation.
5
Reason in conversation
Ask the agent to read a diagram and reason over it — trace dependencies, weigh a change, or propose where new work belongs. Every edit it makes is validated and synced back into the tool.
Try asking your agent
▸ The MCP examples assume the contextgraph server is connected — see Setting up your MCP. The paste-a-link examples need no setup.
With the MCP server connected
claude "List my ContextGraph diagrams, then read the Sales map and tell me what depends on the Payments service"claude "In my Orders diagram, add a Notifications component and wire Orders → Notifications"Or paste a public link — no MCP needed
claude "Read https://ctxgraphai.com/v/ab12cd34.md and list every downstream service of Checkout"