Drive Hilla from any AI.
Hilla runs an MCP server. Any AI that speaks Model Context Protocol — Claude, Cursor, ChatGPT, Windsurf — can read and write to your board directly. The plan lives in the same loop as the code.
What MCP is
The Model Context Protocol is an open standard for connecting AI assistants to external services. Tools like Claude and Cursor implement it natively.
Hilla's MCP server exposes your board as a set of tool calls the AI can invoke. Read state, create cards, update tasks, run agents. No copy-paste, no context loss.
What you can do
Hands-free planning, in your existing AI.
Read your board from any chat
Ask Claude “what's blocked in sprint 2” and it answers from the live state of your board.
Create cards from prompts
Drop a paragraph of context into Cursor and watch tasks materialize on the canvas.
Update task status conversationally
“Mark the auth refactor done and move dependent cards into review.” One sentence, the agent does it.
Trigger agents from your IDE
Cards on the canvas, executed by an AI in your editor. Planning and shipping on the same loop.
Install
One config block. Any client.
Generate a token in your settings, paste it in your AI client's MCP config.
{
"mcpServers": {
"hilla": {
"type": "sse",
"url": "https://api.hilla.ai/mcp",
"headers": {
"Authorization": "Bearer play_YOUR_TOKEN"
}
}
}
}Use cases
Three ways teams already use it.
- Cursor
Plan, build, ship — same window.
Open Cursor on your repo. Ask it to look at the active sprint. It pulls cards, blocks, and dependencies from Hilla. It writes the code, opens the PR, marks the card done. Your canvas updates while you're still in the editor.
- Claude Desktop
Your project, in conversation.
Open Claude. Ask “what should I work on today?” Claude reads your board, factors in priority and dependencies, picks the next card. Says yes, the card moves to In Progress. Move on.
- ChatGPT
Plan from a paragraph.
Paste a project brief into ChatGPT. It calls Hilla's MCP, creates the project, generates the cards, sets the dependencies. You go to hilla.ai, the board is already laid out.
Full tool list, auth flow, rate limits — in the docs.
Read the MCP docs →