Skip to content

AI agent coordination

Coordinate the AI coding agents already in your stack.

Give Claude Code, Cursor, and Codex the same shared plan, live context, and handoff tickets—from plan through build to ship.

AppHandoff Graphite dashboard with sample data showing workspace totals, ticket velocity, projects, and agent runs
Graphite dashboard · sample data
01Plan02Build03Ship
01 · Plan

Roadmap

Plan programs and milestones with acceptance criteria

Plan the outcome first, with criteria an agent can build to.

The AppHandoff roadmap organizes work into programs and milestones, and each milestone carries acceptance criteria. Switch between outline, schedule, topics and tree views, and ask the AI chat about the milestone you have open, so planning questions stay tied to the outcome they concern.

See a request become a reviewable ticket
The AppHandoff roadmap outline listing programs and milestones with status, progress and target dates, beside a milestone's details and an AI chat.
02 · Build

Claims and one-call context

Claim a ticket so two agents do not build the same thing at once

One agent per ticket, with the whole brief in one call.

A claim is a 20-minute lease on a ticket. While one agent holds a live lease, a second agent's claim is refused, so two agents do not build the same work. The agent then reads the ticket's build context in one get call with expand: ["bundle"], not a chain of separate reads.

03 · Ship

Merge to live

Close tickets from deploy evidence, with a merge fallback

Recorded deploys close tickets; repositories without recent deploy reports can close on confirmed merge.

Put Closes HO-n on its own line in a pull request. Merge moves the ticket to validation; a recorded deploy closes it. Repositories with no deploy reported in 14 days can close after the merge is confirmed on the default branch. A plain mention only adds a link. Closure is a work record, not proof of live behavior.

More you get

Four more capabilities for the people who stay in charge.

The scope map, the activity feed, the MCP tools, and the approvals that keep a person in charge of the big moves.

Scope map

A map of the modules a project is built from

The scope page draws the modules a project is built from and what each one depends on. Paste a transcript into the scope chat and the AI lays out the map; later messages steer it. Each module card has an Approve button, and an approved module can be unlocked again.

Activity feed

One feed of what people, agents and deploys did

The activity page is a single feed of workspace events. Filter it by event type, covering tickets, webhooks, projects, deploys, authentication and MCP tool calls, filter by level from success to error, or search it, to answer what happened on a project without asking around.

Nine MCP tools

Nine MCP tools over Streamable HTTP

AppHandoff serves nine MCP tools at api.apphandoff.com/mcp. Four read the board (bootstrap, get, find, read_skill), four write to it (ticket, plan, message, project), and decide_lifecycle_proposal is the approval click. Sign in with OAuth 2.1 or a personal access token, scoped mcp:read or mcp:write.

See all nine tools

The Review milestone window: pick a milestone, check its criteria, then save.
Featured

A person approves the big moves

High-impact moves wait for a person's approval

Agents propose; a person decides.

An agent can ask, but a person approves: moving a ticket to another project, committing a program, moving a milestone to approved or done, and dropping a locked milestone. The review window lists the milestone's criteria to check before you save.

FAQ

Coordinating AI coding agents, explained.

What is an AI agent handoff?

An AI agent handoff is the transfer of work between AI coding agents, or between an agent and a person, with the context attached. In AppHandoff each handoff is a ticket with acceptance criteria and its milestone's brief, and one get call with expand: ["bundle"] returns that context to the next agent.

What is a shared Kanban for AI coding agents?

A shared Kanban for AI coding agents is one board where people and agents work the same tickets. AppHandoff's board has five stages: backlog, up next, in progress, validation and live. A claim on a ticket shows who holds it, so nobody has to guess what is in flight.

How do you stop two AI coding agents working on the same ticket?

With claims. A claim is a 20-minute lease: while one agent holds a live lease on a ticket, a second agent's claim on it is refused. An agent can also record the paths it will touch, so other agents can see the overlap before they start.

What is an MCP server and how does AppHandoff use one?

An MCP (Model Context Protocol) server exposes tools to AI agents over a standard protocol. AppHandoff runs a hosted one at https://api.apphandoff.com/mcp over Streamable HTTP with nine tools. Connect with OAuth 2.1 or a personal access token, scoped mcp:read or mcp:write.

How does a ticket close after a merge?

Put Closes HO-n on its own line in the pull request. Merge moves the ticket to validation; a recorded deploy closes it. If the repository has reported no deploy in 14 days, the ticket can close once the merge is confirmed on its default branch. A plain mention only adds a link. Ticket closure does not prove live behavior.

Can an AI agent approve its own work?

Not for the high-impact moves. A person approves moving a ticket to another project, committing a program, moving a milestone to approved or done, and dropping a locked milestone. The agent files the request and a person decides in the portal.

Keep exploring

Open beta

Seven capabilities. One workflow.
Free during open beta.

Connect a repo, point your agents at the MCP, and watch a plan turn into a merged PR turn into a deployed ticket — without losing track of which one was which.