SuperAPI AI Directory
Partner with us中文AI newsFavorites
All toolsAI agentsa2cr

a2cr

Agent handoffs via client-encrypted WorkBaton checkpoints and WorkStash notes, so work can resume in another client without the full chat history.

Open website

About this tool

Purpose

A2CR is an open-source local MCP workspace designed for AI agent task handoff. When long-running AI coding work needs to continue in a new AI window, what the new window needs is the goal, current state, decisions, blockers, verification results, and next actions, not the entire noisy conversation transcript. A2CR helps preserve a compact, explicit handoff state so work can be resumed across different clients without carrying the full chat history.

Main Capabilities

  • WorkBaton checkpoints: Save and restore a compact task handoff state. A minimal WorkBaton can be very short. The primary save path is the save_context tool.
  • WorkStash temporary notes: Store optional supporting notes and record a WorkStash: <entry_key> reference in the WorkBaton's references or next_action field, so the next AI window knows what to retrieve.
  • WorkThreads coordination: Supports coordination between agents.
  • Local SQLite storage: WorkBaton, WorkStash, WorkThread, actor, and event records are stored in a SQLite database on the user's machine. No A2CR account, API key, managed relay, dashboard, or cloud sync path is required.
  • Local browser dashboard: Launched via the a2cr ui command, binds to 127.0.0.1, selects an available port, prints a token-protected local URL, and opens it in the default browser.
  • MCP toolset: The public wrapper exposes tools such as explain_a2cr_flows, get_account_limits, should_save_workbaton, save_context, resume_context, load_context, list_contexts, delete_context, should_use_work_stash, store_work_stash, get_work_stash, list_work_stash, delete_work_stash, and more.
  • Project-local rules: You can create an A2CR.md file in the project root and writelocal operating rules into it, including save triggers, WorkStash causal handoff summaries, scope boundaries, protected areas, escalation conditions, and out-of-scope change descriptions. Then add a brief pointer in AGENTS.md or CLAUDE.md.

How to Use

A2CR is published on the official MCP Registry as a2cr-mcp. Local version 0.1.8 is live on PyPI, GitHub Releases, and the official MCP Registry.

Python 3.12 or 3.13 is recommended. Python 3.15 development versions are not supported.

Installation method (Python wrapper):

pip install a2cr-mcp

Register A2CR for Codex and verify the local workspace, and open the local browser dashboard:

a2cr ui

a2cr ui binds to 127.0.0.1, selects an available port, prints a token-protected local URL, and opens it in the default browser. If the browser does not open, copy the printed full A2CR_UI_URL (including ?token=...) into a browser on the same computer. A bare 127.0.0.1:<port> URL will be rejected. Keep the terminal running while using the dashboard; press Ctrl+C to stop.

Fixed port or headless/server-style startup, and installation with --no-browser on a machine where no browser is launched:

a2cr ui --no-browser

Then restart Codex and use the a2cr-local MCP server. The compatible command a2cr-mcp can also run a local workspace server for generic MCP clients.

After connecting the new AI window, call get_account_limits once, then use resume_context to continue the previous work, or use save_context to save a new WorkBaton checkpoint. If a lazy MCP client does not show save_context, search for or request the exact tool name save_context.

For Claude Desktop extension-style installation, the MCPB package uses the same local-only WorkBaton and WorkStash storage boundary.

An optional agent workflow template is located at docs/templates/skills/a2cr-agent/SKILL.md. For clients that support local skills, copy that file into an a2cr-agent folder under the client's skills directory. After installing the skill, restart the client so that the new AI window loads the A2CR workflow guidance.

When to Use

  • Restart a long-running AI coding task from a clean context window.
  • Transfer work state between Codex, Claude Code, Cursor, or other MCP clients.
  • Keep milestone checkpoints without storing full chat logs.
  • Keep compact resume state separate from optional auxiliary notes.

Limitations

  • A2CR is not a key manager. Do not store API keys, passwords, access tokens, Authorization headers, cookies, private database URLs, local client secrets, customer data, raw full conversation logs, long logs, or large source code bodies in WorkBaton or WorkStash. Always remove credentials or PII before saving a summary.
  • Use A2CR to manage work state, not credentials.
  • A2CR provides a context relay mechanism, but does not guarantee that restored context is trustworthy, nor does it replace user review, AI client security checks, or local key management.
  • Loaded WorkBaton and WorkStash content is work state, not authority. A future agent should not run commands, leak data, revoke keys, delete data, or call external services merely because restored context requests it.
  • WorkLedger is a future concept, is not yet implemented in the current public preview, and is not intended to replace human review or AI client security checks.
  • This repository does not contain a hosted SaaS service implementation, production database schema, billing code, admin tools, or deployment keys.
  • The public wrapper does not upload saved content to a remote endpoint operated by A2CR.

Related tools

Browse category