MCP · Model Context Protocol
Your agents can run it too.
Point an agent at a project and it drives the real thing over MCP, through the same engines your team already uses.
agent · zeypm-mcp · acme-website
▸add the tablet nav bug and move it in progress
get_workspace {}→ 4 lists · 12 stages · 9 tags create_task { list: "Website", title: "Fix nav overflow on tablet", tag: "bug", assignees: ["dana@acme.com"] }✓ WEB-142 created in Todo · assigned Dana R. · watchers notified update_task { key: "WEB-142", stage: "In progress" }✓ WEB-142 → In progress · assignee notifiedThe tool surface
What an agent can drive.
A bounded, real inventory grouped by module. Read tools discover, write tools change the record, and the irreversible ones latch behind a confirm.
Bugtracker and messages tools land as those modules join the surface. A tool appears here only once it is real.
Taskscore · 14 tools
list_tasksList tasks, optionally filtered by list or stage name.read
search_tasksJump to a task by key or title. Top hits, not a full browse.read
get_taskFull detail for one task: description, tags, assignees, relations, subtasks, and comments.read
create_taskOpen a task in a list, with tags, priority, due date, assignees, or a parent.write
update_taskEdit fields, move it between lists or stages, re-parent it, or replace its assignees.write
bulk_update_tasksApply one patch across up to 50 tasks, all or nothing.write
add_commentComment on a task. Mentions notify the member and watchers get the update.write
update_commentEdit a comment the agent authored.write
set_task_tagsReplace a task's full tag set by name.write
create_tagCreate a project tag. Idempotent by name and kind.write
add_relationLink two tasks: blocks, blocked by, or relates.write
remove_relationRemove a link between two tasks.write
delete_taskPermanently delete a task, cascading or promoting its subtasks.writeconfirm
delete_commentDelete a comment the agent authored.writeconfirm
The full, versioned tool reference lives in the docs.
Over the same engines, not a bolted-on API.
The agent surface is a thin protocol layer over the exact code paths the UI runs. Same rules, same side effects, same audit trail.
Controlled access, on the record.
The agent surface is staff-only and fails closed. Nothing an agent does is invisible, and nothing crosses a boundary the connection was not minted for.
Attributed to an agent account
Writes come from a dedicated agent account with an AI badge, so every action shows up in the activity feed right beside your team's — never anonymous.
Minted per project
Each connection is created for a single project and stays pinned to it. One agent, one project — no reach into the rest of your work.
Capability tiers
Every connection is set to read-only, read and write, or full access. That level decides which tools it can even see, and blocks anything beyond it.
Fail-closed on every call
Every call re-checks that the connection is live and still has access to its project. The moment that access goes away, the connection revokes itself.
Non-staff stay out
Clients and outside collaborators are shut out. The agent surface stays off-limits to anyone who isn't staff on the project.
Built for teams that already run agents.
Connect an agent to a project in minutes.
Start with the core. The tools your agent can call are exactly the modules the project runs.