Run architecture mapping headless, from Claude Code, Codex or CI
Usually a platform engineer or an architect who automates the team's tools
Map, query and manage your architecture from an AI client, a schedule or your own systems, through two plugins and an MCP server.
The problem
Architecture tools assume someone opens them. The map is redrawn when a person remembers to, the impact question waits until someone logs in, and the work it leads to is typed again into the tracker. Meanwhile the team works in a terminal, an AI client and a pipeline, and a tool they have to visit falls behind them.
They need the same map, the same questions and the same work items where they already work: in their agent, on a schedule, and inside their own systems.
How it goes
The same steps the product’s playbook walks. Each one shows the screen at that moment.
Connect once
Sign in once in the browser, or generate a key in the web app. From then on the workspace plugin and any MCP client use a token from the environment, scoped to read, or to read and write. The code plugin keeps its own credential for each repository.
Map each repository on a schedule
The code plugin analyses a repository and syncs its board. Schedule the analysis in your AI client's desktop app, and each run refreshes the boards whose code changed and is recorded on the board, including a run that stopped and why.
Ask the architecture from your agent
In Claude Code or Codex, the workspace plugin answers from the boards. Asked what depends on the event bus, a run with nobody watching traced 23 elements across three apps, 14 of them reached only through code no board draws.
Manage the work from the same session
List the queue, author a work item anchored to the elements it changes, and hand it to the developer whose session builds it. Headless, the plugin makes the choices it would have asked about and lists each one in its report.
Plug in your own systems
Your own tools need no plugin. Any MCP client with a token calls the same server: trace what a change reaches, read the week's changes, list the open work. Make your script, bot or pipeline step that client, starting with a read token.
- Done
End proofDone when the boards refresh on a schedule, and your agents and systems ask and act on them without anyone opening the web app.
Works with
- AI clients
- Claude Code and Codex run both plugins; Cursor runs the code plugin. Any other MCP client reaches the same server.
- Credentials
- A device sign-in or a key from the web app. In CI, the workspace token and each repository's binding credential come from environment variables.
- MCP server
- One URL with a bearer token: boards, impact and dependency traces, work items and epics, playbooks, insights and the change diary.
- Schedules
- A local routine in the Claude desktop app or a scheduled task in Codex runs the code plugin's analysis on your machine.
Where it stops
- Creating a workspace, generating the first key, publishing a board, single sign-on and billing are done in the web app.
- ProvenMap runs no scheduler of its own. Scheduled analysis runs in your AI client's desktop app while the computer is awake, and not yet in a Claude Code cloud routine.
- Setting up a new landscape is a conversation: the workspace plugin asks as it lays out your systems. It is not meant to run unattended.
- A read and write token changes the workspace. Board edits can be undone from its working copy; work items and insights are permanent. Writes are best left to the workspace plugin.
Why not a diagram, or a code browser
A diagram shows
A file someone opens and redraws by hand. No pipeline can ask it a question, and nothing in it updates itself.
A code browser shows
Search and navigation for a person at a screen, one repository at a time, with no interface an agent or a pipeline can ask what depends on a contract.
ProvenMap shows
The map, the impact traces and the work items behind one MCP server and two plugins, so an agent, a schedule or your own system runs them, and the web app is where people read the result.
Walk through it
- The docs walkthrough of this job is being written.
The features behind it
Start with one system
Map a real system and see the architecture around it.
Open one repository and run the code plugin. No migration, no rollout, no diagramming.
No credit card · Analysis runs locally in your own session · Self-hosted available
See the four commands to your first board →
Or for writing on architecture that stays connected across systems.