Windy.
Sign inStart free

How it works

Design, plan, and steer agent-written code from one source of truth.

Windy is a shared workspace for specs, diagrams, and execution plans. Humans review it in the web app. Coding agents read and write it over project-scoped MCP.

Start freeConnect your agent

What Windy is

A source of truth with two doors: web for humans, MCP for agents.

Windy is not another coding agent. It is the shared design and planning layer used by the coding agents you already have. Humans use the web app to create, organize, edit, and review project intent. Agents use MCP to read and write the same project-scoped source of truth.

  • Project-scoped MCP endpoint.
  • Revocable access.
  • Agents see only the connected project.
  • Humans remain in control of review and direction.

Workflow

The design-first loop.

01

Capture the design

Start with the intent: what the system should do, how the parts fit together, and what constraints the agent must respect.SpecificationsArchitecture notesDiagramsContractsRequirementsDecisionsConstraints

02

Expand the design

Ask your coding agent to create or refine specs, diagrams, and structured docs through Windy MCP. The agent can help write the design, but you stay the architect and reviewer.

03

Create a plan

When the change is too large for one prompt, break it into ordered tasks with objectives, deliverables, dependencies, acceptance criteria, and prompts.

04

Build

The coding agent reads the relevant docs and tasks from Windy before editing code. It builds to the design instead of guessing from a single prompt.

05

Review and refine

Executions record progress. As behavior changes, humans or agents update the docs so the source of truth stays aligned with the code.

Two instruments, one source of truth

Docs capture intent. Plans sequence work.

Docs — the thinking behind the code

Docs hold the project intent your coding agents build from: specs, architecture, diagrams, schemas, contracts, requirements, and decisions.

Supported content
Markdown specsArchitecture notesDiagramsMermaidJSONYAMLAPI contractsDecision recordsRequirements
  • Shared understanding
  • Better grounding
  • Reduced hallucination from missing context
  • Easier review and maintenance

Plans — reviewable progress for large agent work

Plans turn large changes into ordered tasks the agent can execute one step at a time, with objectives, deliverables, dependencies, acceptance criteria, prompts, and execution logs.

What's in a plan
Ordered tasksObjectivesDeliverablesDependenciesAcceptance criteriaAgent promptsExecution logs
  • Predictable execution
  • Safer migrations
  • Reviewable steps
  • Audit trail

Docs tell the agent what should be true. Plans tell the agent how to get there.

See it in practice

What using Windy feels like.

No new tool, no copy-paste — drive Windy from Claude Code, Codex, or any MCP-aware agent: design it, plan it, then build from it, all in one session.

Boundaries

Windy keeps the human as architect.

Windy improves coding agents by giving them design context and task structure. The developer still decides what should be built, reviews the design, and owns the result.

  • Not a no-code builder.
  • Not a replacement for developers.
  • Not another coding agent.
  • Does not execute code.
  • Does not promise autonomous software creation.

Security & control

Scoped to the project. Controlled by you.

  • Project-scoped MCP endpoints.
  • Revocable access.
  • Project isolation.
  • Agents only see the connected project.
  • Humans review and own the source of truth.

Give your coding agents the design before they write the code.

Start with one project, connect your agent over MCP, and let it build from a source of truth instead of a guess.

Start freeRead getting started

windylogic.ai