Spec dumps are amateur hour.

You need iterative development and repeatable tests to use AI effectively.

Work one story at a time. The agent reads the acceptance criteria, writes BDD tests and code. Run the tests on every commit. Move to the next story.

Start a trial Read the docs

Terminal

claude mcp add --transport http mainline \
  https://mainline.dev/mcp/YOUR_TEAM_ID \
  --header "Authorization: Bearer YOUR_MCP_TOKEN"

Generate an MCP token in team settings. Read-only. Works solo or with a team.

LIST_MAPS

Every map for your team. The agent starts here.

GET_MAP

Activities, tasks, stories, status, release slices, personas. What's being built for whom and where each story sits.

GET_STORY

Customer can save a shipping address
Given a logged-in customer at checkout
When they save an address
Then it appears pre-filled on their next order

The agent sees your ACs and can make them executable.

Mainline already stores your acceptance criteria as given/when/then scenarios. The MCP hands them to the agent. No copy-paste, no re-explaining what you already wrote down.

PRIVACY

The MCP token is read-only. The agent cannot modify your map, create stories, or release anything.
No personal information is exposed, story assignments are aggregate.
MCP token and deploy token are separate. Use both or either.

Your acceptance criteria, in your agent's context.
Free to connect, read-only, two minutes to set up.

Start a trial Read the docs