Plainwork
Sign inStart free
For agents

Your workspace.
Their interface.

Most tools treat agents like a chat bubble. Plainwork treats them like a teammate, with an identity, an audit trail, and a place to propose work before it lands.

Start freeRead the docs

Agents do not need a chat window. They need a workspace.

Giving an agent a chat window is like giving a new hire a Slack DM and no workspace access. They cannot see your work. They cannot file theirs. They cannot reference what is already there. Plainwork is the doc tool, the Git history, and the tracker, all in one place. And it works the way they think.

Three primitives. Everything falls out of them.

01

Identity

Each agent gets its own API key, display name, colour, and avatar. Everything it touches is attributed. You filter by agent the same way you filter by human.

02

Proposal gate

By default, agent writes land in a pending inbox. Approve in one click. Edit first if you want. Reject with a reason. Trust an agent? Flip auto-approve.

03

MCP transport

Plainwork speaks MCP natively. Point Claude Desktop, Cursor, Claude Code, or anything else at your workspace. Two lines of config.

The tools your agents get.

Same primitives humans use. Exposed as typed tool calls.

list_collectionsEnumerate every collection the agent can see.
list_viewsList views on a collection.
read_pageRead a note, properties and body.
query_pagesFilter, sort, group. Same as a view.
search_pagesHybrid keyword + semantic search.
create_pagePropose a new note.
update_pagePropose an edit.
delete_pagePropose a delete.
move_itemPropose a move between collections or folders.

Writes route through the approval gate unless the agent is set to auto-approve.

Approve. Edit. Reject. Move on.

The inbox is a side panel, one click from any page. Cards show exactly what changed. Hit Approve, or open the editor and tweak the content before it lands.

Pending actions (3)

Claude, Cursor, Claude Code

Create note. "Competitive teardown: Acme v4"

proposed by Claude · 2 min ago

ApproveEditReject

Update note. "Launch checklist". 3 blocks changed

proposed by Cursor · 5 min ago

ApproveEditReject

Move note. "Press list" to Archive

proposed by Claude Code · 7 min ago

ApproveEditReject

Wire it up in 60 seconds.

Here is the Claude Desktop config. Cursor, Claude Code, and anything else that speaks MCP work the same way.

{
    "mcpServers": {
      "plainwork": {
        "command": "npx",
        "args": ["-y", "@plainwork/mcp"],
        "env": {
          "PLAINWORK_API_KEY": "pw_agent_..."
        }
      }
    }
  }
  • Generate a key in Settings, then Agents.
  • Keys are shown once. Store them somewhere safe.
  • Scope a key to a single collection or the whole workspace.

Your agents. Your rules.

Attribution.

Every action, every note, tagged with the agent that made it.

Approval gate.

On by default for untrusted agents. Off per-agent when you are ready.

Revoke instantly.

Kill a key from Settings. The agent stops mid-sentence.

Audit trail.

Who approved what, when, with reasons. Queryable.

Status

Agent identity, attribution, and the pending inbox are live today. The Plainwork MCP server is rolling out soon. Join the early access list to get a key when it ships.

Bring your agents home.

Create your workspace