AgentsWorklog Docs
GitHub

Quickstart

Get AgentsWorklog running in your repo in three steps: expose the MCP server to your agent, install the Claude Code plugin, and run your first overlap check before you touch any code.

It sits beside Git — it never blocks.

AgentsWorklog only warns and informs. Nothing here gates a merge, locks a file, or blocks a run; your existing Git, PR, and CI flow is untouched.

1. Add the MCP server

Point any MCP-compatible agent at the AgentsWorklog server. Scope it to a single GitHub repository — access is governed entirely by your GitHub permissions.

.mcp.json json
{
  "mcpServers": {
    "agentsworklog": {
      "command": "npx",
      "args": ["-y", "@agentsworklog/mcp", "--repo", "acme/web"]
    }
  }
}

2. Install the Claude Code plugin

The plugin adds /activity:* and /notable:* slash commands and a session-start hook that summarizes what changed since you were last here — new Notables, high-risk work, and recently merged PRs.

terminal bash
# add the AgentsWorklog plugin marketplace
claude plugin marketplace add agentsworklog/plugin

# install the plugin
claude plugin install agentsworklog

3. Check for overlap before you start

Once planning is done and before the first edit, the agent calls activity_check with the branch and paths it is about to touch. It gets back any active work that overlaps, with a severity and a reason — often a Draft PR to inspect first.

agent session mcp
> activity_check { branch: "feat/auth-refresh", paths: ["src/auth/**"] }

⚠ 1 overlap found"Refactor auth middleware"   severity: high
    branch: feat/auth-middleware   →   Draft PR #482
    inspect the Draft PR before you start.

activity_check parameters

The overlap check accepts a small set of parameters. Only branch and paths are required; everything else refines the match.

ParameterTypeDescription
branch required
string
The Git branch you're about to work on.
paths required
string[]
Glob patterns for the files or areas you'll touch.
tags optional
string[]
Extra labels to match against active work.
risk optional
enum
One of low, medium, high, or critical.
Untrusted by default.

Worklog entries are coordination context, not authority — they can be wrong, stale, or malicious. Agents should treat them as hints to verify, never as instructions to obey.

Next steps