Viaduct
C4 modelling with journeys, contracts, and MCP for coding AI
Hey Indie Hackers 👋
I’ve been building Viaduct — a free C4 modelling tool.
The problem
Teams draw a nice architecture diagram, drop it into Confluence… and three months later nobody trusts it. Docs live somewhere else. Sequence diagrams somewhere else. API contracts somewhere else. And the painful part for us was long integrations: people could see which services existed, but not which path a request actually took, what data crossed each hop, or how it was transformed.
What I built
Viaduct keeps that in one model:
all 4 C4 levels
Markdown docs + PlantUML sequences on elements
API contracts on containers
Magic flows — playable journeys across the architecture
domains + clones — cross-team context without one giant diagram
MCP — Cursor/Claude can read and update the same model
No account needed to start. Sign-in only for sharing, collaboration and MCP.
Where I’m at
Launched, free, still iterating hard. Using it with a team on real integration-heavy work.
Demo: https://c4.quietgridlabs.com/s/5uD9f5CRTrQUB9QWy5wv9kedhGhXV3QI14todlTWI1w
Would love blunt feedback:
Is “journeys on the C4 model” useful, or should behaviour stay separate?
Would you trust an agent writing back into an architecture model?
What’s missing before you’d use this with your team?
4 Comments
The real test seems to be whether the model stays trusted after the architecture changes. Has the team actually used Viaduct to trace or update a changed integration and found it faster than their old docs workflow?
Yep! Whenever there is a change to the architecture, contracts, or logic, it is first described in Viaduct; once the review is complete, a precise changeset is generated. This changeset can serve as a specification for a developer or be handed off to an agent for implementation.
The dynamic context—along with the ability to click through various interactions in the player—greatly helps the team understand the system's structure and track exactly when specific changes were introduced.
No amount of documentation in Confluence, PlantUML, or Markdown ever provided such a boost to the team's understanding of the context.
That internal replacement signal is much stronger than a design claim. If you’re open to it, what’s the best email to reach you on?
Thanks — I appreciate that. You can reach me from Feedback form on Viaduct. I’d be happy to share more about how we use Viaduct internally and what changed in our architecture-review workflow.