Use the conversation and coding workflow you already know.
Durable context for AI-built software
Give your coding agent
a real contract.
Bring Codex, Claude Code, or another MCP-compatible coding agent. Benchwright brings the MCP connectionthat turns your product decisions into a durable contract. You stay in the coding workflow you already use.
Durable product context connected to your agent.
How you interact
- 01Describe
Tell your agent what you are building.
- 02Review
Challenge the proposed product structure.
- 03Answer
Make the decisions only you should make.
- 04Approve
Let the agent build from the shared contract.
One workflow, three roles
You decide. Your agent works. Benchwright keeps the contract.
Work where you already work.
Describe a product or change in your coding-agent conversation. Review proposals, answer focused questions, and approve the resulting contract.
Benchwright joins through MCP.
We provide the connection between your agent and durable product context. There is no separate Benchwright chat or second workflow to manage.
Keep approved truth durable.
Benchwright stores features, system boundaries, behavioral claims, and exact test identities, then returns focused context for each implementation task.
Inside your coding agent
See how you work with Benchwright.
Follow a real CAD Searcher interaction from the first prompt, through review and focused product decisions, to an approved contract the coding agent can build from.
Your existing coding-agent conversation
Start in the coding agent you already use.
There is no Benchwright chat to open. Describe the product or change to your coding agent in ordinary language; the agent uses Benchwright through MCP to turn that brief into durable product context.
I'm building a system that allows a user to take pictures of an object and search a CAD database for that object. We're using the ABC Dataset for testing, DINO as our vision embedder, and YOLO as our object cropper.
Benchwright is connected. I'll turn this into a product proposal, call out the decisions that still need your judgment, and bring the contract back for review.
You never leave this conversation.
The agent extracts the durable outcome from your prompt.
Named technology is preserved without being mistaken for a complete product contract.
The agent can now ask what the technical brief leaves unresolved.
You stay in your coding agent. MCP is the connection; Benchwright is the durable record behind it.
This conversation happens inside your coding agent. Benchwright is connected through MCP; there is no separate Benchwright chat to manage. Example records are simplified for clarity.
One answer, precisely carried forward
You answer the agent. Benchwright preserves the decision.
The coding agent translates a short human answer into behavior, ownership, and proof. Benchwright preserves that approved context while the agent builds.
“When multiple objects are detected, the user selects the crop before search begins.”
Behind the connection
A durable graph behind every agent interaction.
MCP gives the coding agent access to structured product truth: Products decompose into Features, Claims belong to responsible Systems, and Tests keep behavior tied to executable proof.
Product
The durable intent for what your organization ships.
decomposes into durable FeaturesFeature
A persistent, user-facing capability confirmed by people.
is defined by Claims owned by supporting SystemsClaim
A falsifiable statement of observable behavior.
belongs to a System and is verified by TestsSystem
The architectural boundary responsible for a Claim.
limits the code and context an AI must inspectTest
Executable evidence for one or more Claims.
maps one-to-one to a test function in codeStory
A declarative handoff that groups Claims for a change.
focuses implementation without replacing durable truthYour agent, connected
Bring your coding agent. We'll bring the durable context.
Work in the coding workflow you already use. Benchwright connects through MCP and keeps intent, responsibility, and evidence aligned.
Request early access