dlvr.one gives your coding agents, for example Claude Code, a real place to work: the same projects, user stories and use cases your team uses. Connect them through an MCP server, an agent API and signed webhooks. People approve at the gates.
Invite-only. Early access. Bring your own agent.
URL https://<your-instance>/mcp
Header X-API-Key: <your-mcp-key>
Scope mcpAgents do not get a side channel. They use the same projects and use cases as everyone else.
Any MCP client can read and edit the backlog as the key's user over streamable HTTP. A full toolset covers spaces, projects, epics, user stories, use cases and the rough backlog, including the whole review lifecycle.
A defined REST contract lets an external agent drive a use case end to end without a human clicking.
dlvr.one tells your agent when something needs it, and your receiver can verify the sender with an HMAC-SHA256 signature. Register HTTPS targets for the whole instance or per space.
Webhooks carry agent-originated events. Everything else is available to read through MCP and REST.
dlvr.one does not run an AI model itself and ships no agent. You connect the one you already use.
Keys act with exactly their owner's permissions, so what an agent can see is what its person can see.
Personal API keys, scoped mcp or agent. The raw key is shown once; only a hash is stored.
Revoke a key and it stops working on the next request.
Per-project visibility applies to the web app, MCP and the agent API alike. Hidden items look like missing ones.
Choose the executor per use case: your agent or a person. Teams without an agent can run the manual path.
Every use case moves through one pipeline. The agent cannot ship on its own.
Raw requirements become a technical specification and a test plan, written by your agent or by a person.
Approved work is implemented. If the agent is unsure it raises an intervention and waits for an answer.
A validation package is reviewed at the gate. A human approves it, or sends it back to Forging with a note.
Humans decide at three checkpoints: approve out of Staged, answer an intervention, approve or reject at the gate.
No agent required to get value. The structure is the classic one.
Use cases get readable keys, UC-12 for work and BUG-13 for bugs.
Group work by client or product, with members, status and dates.
Capture raw notes first, then convert a note into one or many user stories.
Requirements, priority, assignee, epic and labels, with a backlog view and a My work overview.
See what waits for the agent and what waits for a human, at a glance.
Project labels and space-wide labels to organise user stories.
Plan dates, spot overdue work, and record the delivery date when it is done.
One organization per instance, and every customer gets their own deployment and database.
There is no public sign-up. The owner invites people by email with a one-time set-password link.
People see a project and everything in it only as a member. Owners and admins see all.
Owner, admin, manager, user and client roles.
Let customers file requests and follow a read-only board, without exposing prices, work logs or technical specs.
We are inviting teams one by one. Write to us and tell us how you work with AI agents today.
Get instanceOr just talk to us: [email protected]