One source of truth, every agent.
Specs, architecture, diagrams, proposals, task lists, and execution history live in one workspace that every connected agent reads and writes.
Solutions
Works with your agents
Windy is agent-agnostic. Connect Claude Code, Codex, Cursor, OpenCode, or any MCP-aware coding agent to the same workspace. Each person gets their own attributed per-(user, project) MCP endpoint, so every change an agent makes is traceable to a human.
One workspace, every agent
Claude CodeCodexCursorOpenCodeMCP-aware coding agents
Bring the agent you already use — Windy doesn’t pick one for you.
Windy is not tied to a single coding agent. It exposes your project’s design, proposals, task lists, and execution history over the Model Context Protocol, so any MCP-aware agent can read and write the same workspace. Switch agents, or let different teammates use different ones, without losing the plan.
- Claude CodePoint it at Windy from CLAUDE.md and it reads the project's specs, proposals, and task lists over MCP before it edits.
- CodexConnect the endpoint and Codex works from the same proposal specs and knowledge base every other agent sees.
- CursorAdd Windy as an MCP server so in-editor edits build from durable project context, not a one-off prompt.
- OpenCodeWire the endpoint into your config and OpenCode joins the same reviewed workflow as the rest of the team.
- Any MCP-aware agentWindy speaks the Model Context Protocol, so future and in-house agents connect the same way — no bespoke integration.
Windy is not another coding agent. It is the workspace your agents work in — the shared design, plan, and review your agents build from, whichever one you run.
Endpoints & attribution
Each person connects on their own endpoint, so every change has a name on it.
Windy’s MCP endpoint is scoped per (user, project), not per project. Every teammate connects their agent on their own revocable endpoint, capped per user. Because an agent always runs on a person’s endpoint, every proposal, comment, and merged changeset it makes is attributed to that human.
Per-person endpoints.
Each teammate connects on their own per-(user, project) MCP endpoint — revocable, scoped to the project, and capped per user.
Every change is attributed.
Because an agent runs on a person's endpoint, every proposal, comment, and changeset it makes is traceable back to a named human.
Switch agents without losing context.
Move from Claude Code to Codex to Cursor mid-project and the design, plan, and history stay put — the workspace is the constant.
One reviewed workspace keeps many agents aligned — and keeps everyone accountable.
Connect
Three steps to wire up any agent.
The setup is the same whichever agent you use — add one MCP endpoint, then let the agent read and write the project’s workspace.
01
Create your endpoint
In the Windy web app, generate your per-(user, project) MCP endpoint for the project you want the agent to work in.
02
Add it to your agent
Paste the endpoint into your agent’s MCP configuration — Claude Code, Codex, Cursor, OpenCode, or any MCP-aware tool.
03
Build from the workspace
The agent reads the current proposal, task list, and knowledge base before it edits, and writes its work back for review.
FAQ
Questions, answered.
Windy is agent-agnostic. It works with Claude Code, Codex, Cursor, OpenCode, and any MCP-aware coding agent. Because it speaks the Model Context Protocol, new and in-house agents connect the same way.
No. Different people can use different agents on the same project, and you can switch agents mid-project. They all connect to the same Windy workspace, so the specs, task lists, and execution history stay in one place.
Windy exposes a per-(user, project) MCP endpoint. You add that endpoint to your agent's MCP configuration once, and the agent can then read and write the project's proposals, knowledge base, task lists, and executions over MCP.
Yes. Every teammate gets their own endpoint, bound to their user and scoped to the project, with a per-user cap. Endpoints are revocable, and every change an agent makes is attributed to the person whose endpoint it ran on.
A per-agent memory silo drifts — each agent ends up with a different idea of the project. Windy is one shared, reviewed workspace: agents propose changes, humans review them, and merges are atomic, so the whole team builds from the same truth.
Related
Keep exploring.
Bring the agent you already use. Windy is the workspace it works in.
Connect your coding agent over MCP, keep the design and plan in one place, and know exactly who changed what.
windylogic.ai