DoviloDovilo: the task list your coding agent works out of.
Claude Code, Cursor and any MCP client connect from any machine and see only their own AI tasks.
In short
Dovilo is a to-do app with a separate task list for AI coding agents. Agents connect to dovilo.app/mcp/agents over MCP: 9 tools, no create, no delete, and your personal list is never visible. ChatGPT or Claude can plan the work; every instruction waits for your approval. Pro, $2.49 a month, 15 AI tasks a month.
How an AI task runs
-
Write the task
Or let ChatGPT or Claude propose it.
-
Approve it on your phone
Nothing reaches an agent before you say yes.
-
Your agent works it
Steps and notes land on the task as it runs.
-
It asks, you answer
It stops at a decision; you reply from your phone or chat.
Use case: Dovilo is built with AI Tasks
Dovilo is built this way. It has five projects, Backend, Mobile, Desktop, Edge and Landing, and one Claude Code agent for each. I plan a change in Claude, which proposes AI tasks under the right project. I approve them on my phone; the agent at the desk works them, writes its steps and notes on the task, and stops to ask when a decision is mine. This page went through the Landing project the same way.
Works with any MCP coding client
- Claude Code, Cursor, Codex CLI, Gemini CLI, Windsurf and Claude Desktop work the tasks.
- LangGraph, CI jobs and any MCP client connect with the same URL and key.
Connect a coding agent: a URL and a key
claude mcp add --transport http --scope local dovilo https://dovilo.app/mcp/agents --header "Authorization: Bearer <API_KEY>"Create an agent in the app, copy its key once, and point your client at https://dovilo.app/mcp/agents. Dovilo Desktop writes the config for 7 clients.
A project carries its brief and its documents
Specs, decisions and checklists live in the project. Your coding agent reads them with read_context; a sub-project's agent reads its parent's too.
- Drafted in chat, approved on your phone. ChatGPT or Claude proposes a document or an edit.
- Agents read, never write. No agent tool changes a document.
- Markdown, versioned. Up to 200 KB a document and 50 a project; the last 5 versions are kept.
- Deleted only in the app. No MCP has a delete tool, and a deleted document is gone for good.
What your coding agent can and cannot do
- Its own tasks, nothing else.
- A person writes or approves every task it gets.
- It cannot create or delete tasks. Cancelling leaves the task visible.
- Your coding work stays out of your chat app unless you turn it on.
- Revoke any agent key from your phone, instantly.
9 tools: 3 read, 6 write. No create, no delete
| Tool | Access | Does |
|---|---|---|
list_tasks | read | AI tasks in the scoped projects, with status |
get_task | read | One task in full: description, steps, notes, telemetry, the user's latest reply |
read_context | read | The project brief and documents, and those of its parent project |
update_status | write | queued → running → awaiting-user → done; done-with-errors, error; cancel with a reason (stored as error + cancelledAt, shown as Cancelled) |
update_step | write | Mark a plan step started or finished |
add_step | write | Append a step the agent discovered |
skip_step | write | Skip an unnecessary step |
append_note | write | What it did, decided, could not do |
append_telemetry | write | Model, tokens, cost, duration |
Plan in ChatGPT or Claude, hand it to your agent
With AI task access on for a connection, ChatGPT or Claude can propose AI tasks under your projects, follow their steps and draft your answer when an agent asks. Every proposal and every answer waits for your approval on the phone.
Webhooks and an open-source demo
Signed webhooks
Delivered by the server, retried up to 5 times.
Open-source demo
A LangGraph agent runs the whole loop. Open source.
What it costs
- Pro$2.49 a month, $19.99 a year or $49.99 once; 7-day trial in the app.
- AI tasks15 AI tasks a month, top-ups from $0.99.
- PlansSending a plan is free, approving needs Pro.
Straight answers
Does my agent need Dovilo Desktop running?
No. Agents connect to dovilo.app/mcp/agents directly. Desktop is the Windows and Mac client for AI tasks, keys, briefs and webhooks.
What does the agent see?
The AI tasks of the projects it is assigned to, each project's brief and documents, the brief and documents of a parent project, and your replies. Your personal to-do list is not on the tool surface at all.
Can an agent delete a task?
No. There is no delete tool on either server. An agent can cancel a task with a reason, and the task stays visible.
Can I run it in CI or on a server?
Yes. The same URL and key work from any machine. One key per agent, revocable from your phone.
I used Desktop 1.8. What changed?
1.8's local server is gone. Create an agent in the app, copy its key, and point your client at the remote URL. Steps in the MCP docs.
One engineer, in Istanbul.
No company, no funding. A public release history you can check.