Shared context and human review for AI coding agents

Agents build.
Humans approve.

Today the plan lives in the agent's context window — it dies at compaction, and your first look at the work is a 500-line diff. Taskum serves shared project context to every agent over MCP, and the review gates you turn on are enforced by the server: execution is refused until you approve the plan.
Works with Claude Code, Cursor, Windsurf, and any MCP agent.
taskum.com
The taskum inbox with a design document open for review — the reviewer's Approve button beside it; the agent cannot proceed until they decide
A design doc awaiting your approval. The agent can't proceed past it until you decide — and execution stays refused until the plan after it is approved too.
Today
prompt 500 plausible lines skim merge rewrite, three features later
With Taskum
idea design doc REVIEW plan REVIEW code clean PR
the full workflow — lighter ones exist, down to none
The knowledge compounds. Every issue your agent completes makes the next one better — not because the model improves, but because the project remembers: the docs stay current and the record accumulates.
Every agent reads the same project context — Claude Code, Cursor, Windsurf, Cline. A teammate's agent knows what yours decided. A new teammate is productive on day one.

Connect an agent in a minute.

OAuth in the browser — no API key to mint. Create a project and ask your agent to read the codebase.

Get started Read the docs →