Let your AI agent tend your tree

Aug 2026 · Feature deep-dive · 5 min read

Your AI agent can already write code, draft emails and summarise your week. Why is your task list still something you update by hand?

The LatticeRun MCP connector closes that loop. It exposes your projects as a set of MCP tools over Streamable HTTP, so Claude Code, Kimi, Hermes Agent — or anything that speaks MCP — can read your tree and keep it moving while you watch the bubbles update live.

What an agent can do

latticerun_list_projects — see your projects and their progress at a glance
latticerun_get_tree — read a full task tree before reasoning about it
latticerun_add_task — break a goal into branches and sub-branches
latticerun_toggle_task — check work off (or honestly reopen it)
latticerun_rename_task / latticerun_remove_task — keep the tree tidy
latticerun_set_due_date — deadlines and reminders, set from chat

"Turn my launch checklist into a LatticeRun project" goes from a copy-paste exercise to one sentence.

Project Plan and Project Memory

A founder, an agent, or both can draw and edit a Visual Project Plan flowchart. Plan Mode is optional; approve and release it when ready to create or link executable tasks. Durable decisions, guardrails, architecture, checkpoints and handoffs live in shared, durable Project Memory, with trust labels for successors.

How you connect

Generate a token in the app (profile menu → Agent tokens), then point your client at the connector:

claude mcp add --transport http latticerun https://mcp.latticerun.com/mcp \
  --header "Authorization: Bearer latticerun_your_token_here"

The full guide, including JSON config for Kimi and Hermes, lives on the MCP connector page.

Scoped by design

Handing an agent write access to your plans requires actual restraint, not vibes:

Tokens see what you see. If you're a task-scoped collaborator on a project, your token inherits exactly that scope — nothing more.
Revocable in one click. Tokens are shown once at creation and can be killed anytime from the same menu.
Every change is undoable. Agent edits land in your normal undo history and sync like any other edit. Nothing an agent does is a one-way door.

Why MCP and not "yet another API"

Because the agents you already use speak it natively. No SDK to learn, no auth dance to re-implement — one URL, one header, and your existing tools gain a new skill. The connector is in beta for Solo Pro and Team plans. Take a look at the plans →