How agents work with Treehouse
The two ways an agent can reach a workspace, what it can do there, and how its work shows up for everyone else.
An agent in Treehouse is a teammate with access to the workspace. It reads the same files you do, writes new ones, reorganises folders and answers review comments. Everything it does is versioned, so a bad edit is one restore away, and a clash with someone else's edit leaves a conflicted copy rather than lost work.
Two ways in
| Through a synced folder | Through the MCP server | |
|---|---|---|
| How | The agent works on a folder on your computer that the desktop app or CLI keeps in sync | The agent connects to https://api.trytree.house/api/mcp with an API key |
| Good for | Coding agents with file access: Claude Code, Codex, Cursor, Aider | Agents without your filesystem, agents running elsewhere, and anything that needs comments |
| Setup | Open the agent in the folder | One command or config entry, per client |
| Its changes show as | Yours, because they sync through your device | The agent's, labelled with the key's name |
| Comments | Not visible (comments aren't files) | Read, reply and resolve |
| Creates share links | No | Member links only |
You can use both. A common set-up is Claude Code working in the synced folder for heavy editing, with the Treehouse MCP server also connected so it can pick up review comments.
What an agent can do over MCP
The MCP server gives an agent the same abilities you have in the web app, apart from a few that deliberately need a person:
- Files: list, search by name, read, write, move and delete, including binary files like images and PDFs.
- History: read any earlier version of a file and restore it.
- Organisation: set file and folder icons, and mark things private.
- Comments: list review threads, reply, resolve them, and start new ones.
- Sharing: create and revoke member links.
- Workspaces: with an account-wide key, list your workspaces and create new ones.
What it can't do: create public links (those need a person signed in to the web app), invite or remove members, manage API keys, or change billing. The MCP server reference lists every tool.
How agents know what to do
Two plain files carry a workspace's instructions:
AGENTS.mdat the workspace root is its rulebook: how it's organised, what goes where, what an agent may and may not do. The Treehouse MCP server tells every agent to read it first, and coding agents look for it by convention. Write your AGENTS.md shows what to put in it.- A
README.mdin each folder says what belongs in that folder. It's also the folder's front page in the web app, so people and agents read the same description.
Agents that support skills can also load the Treehouse agent skill, which teaches the general rules of working in any Treehouse workspace. The agent handbook is the same material as a docs page.
What Treehouse doesn't do
Treehouse doesn't run agents. It has no scheduler, no model and no background jobs of its own. If you want an agent to process an inbox every morning, the schedule lives wherever the agent runs: a Claude Code routine, a cron job, a cloud agent. Routines shows a few ways to set that up.
Related
- Connect Claude Code
Give Claude Code access to a workspace through your synced folder, the Treehouse MCP server, or both.
- Manage API keys
Create a key for each agent, see which keys can reach your workspaces, and revoke the ones you no longer need.
- Agent handbook
Written for agents. How to reach a Treehouse workspace, the rules that keep everyone's work safe, and the conventions the web app understands.
Last updated