A2CR

A2CR is an open-source local MCP workspace for AI agent handoffs. It lets Codex, Claude Code, Cursor, and other MCP-capable agents save WorkBaton checkpoints, store temporary WorkStash notes, coordinate through WorkThreads, and resume long coding work from a fresh AI window.

Long AI work usually breaks at the handoff. A fresh AI window needs the goal, current state, decisions, blockers, validation, and next action, but not a whole noisy transcript. A2CR keeps that handoff state compact, explicit, and safer to share between sessions.

Use A2CR when you want to:

• restart a long AI coding task from a clean context window

• pass work state between Codex, Claude Code, Cursor, or another MCP client

• keep milestone checkpoints without storing full chat transcripts

• separate compact resume state from optional supporting notes

Japanese overview | MCP setup | Usage guide | WorkBaton spec | Local mode spec | 0.1.8 release notes

Directory Status

A2CR is published in the official MCP Registry as io.github.a2cr/a2cr-mcp. The local-only 0.1.8 public release is live on PyPI, GitHub Releases, and the official MCP Registry. See the distribution inventory for the current registration and mirror status.

A2CR is also listed and evaluated on the Glama MCP Registry. Downstream directories may temporarily show cached metadata after a release; the official MCP Registry and this repository are the source of truth.

Related MCP server: agent-gate

Local Storage Boundary

A2CR now treats the local workspace as the public product boundary. The public a2cr-mcp wrapper stores WorkBaton, WorkStash, WorkThread, actor, and event records in a SQLite database on the user's machine. It does not require an A2CR account, API key, hosted relay, dashboard, or cloud sync path.

The earlier hosted/SaaS relay path is being retired from the public distribution. New public setup, MCP Registry metadata, and Anthropic Directory submission work should use the local wrapper only.

Quickstart

Choose the local MCP distribution path that matches your AI client:

Keep the Python wrapper version and Node MCPB compatibility version aligned. For the current public release, both the Python wrapper and Node MCPB compatibility version are 0.1.8.

Python wrapper install:

python -m pip install --upgrade a2cr-mcp

Register A2CR for Codex and verify the local workspace:

a2cr init codex --local
a2cr doctor --target local

Open the local browser dashboard:

a2cr ui

a2cr ui binds to 127.0.0.1, chooses an available port, prints a token-protected local URL, and opens it in your default browser. If the browser does not open, copy the full printed A2CR_UI_URL, including ?token=..., into a browser on the same computer. The bare 127.0.0.1:<port> URL is rejected by design. Keep that terminal running while you use the dashboard. Press Ctrl+C to stop it.

For a fixed port or a headless/server-style launch:

a2cr ui --port 50895
a2cr ui --no-browser

Use --no-browser when installing on a machine where the terminal cannot or should not launch a browser; it still prints the full local URL to copy.

Then restart Codex and use the a2cr-local MCP server. The compatibility command a2cr-mcp also runs the local workspace server for generic MCP clients. Start with a compact WorkBaton save/resume. Put larger supporting notes in WorkStash, then record the WorkStash: <entry_key> reference in the WorkBaton references or next_action field so the next AI window knows exactly what to retrieve.

Codex-style TOML:

[mcp_servers."a2cr-local"]
command = "a2cr-local-mcp"
args = []

[mcp_servers."a2cr-local".env]
A2CR_LOCAL_DB = "/optional/path/to/a2cr.db"

Generic MCP JSON:

{
  "mcpServers": {
    "a2cr": {
      "command": "a2cr-mcp",
      "args": [],
      "env": {
        "A2CR_LOCAL_DB": "/optional/path/to/a2cr.db"
      }
    }
  }
}

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

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

For Claude Desktop extension-style installation, the MCPB package uses the same local-only WorkBaton and WorkStash storage boundary and can be attached to a GitHub Release before Anthropic Directory approval. See docs/claude-desktop-mcpb.md.

Local Project Rules

For project-specific A2CR behavior, create A2CR.md in the project root and put the local operating rules there. Use the repository-root A2CR.md as a starter template. Then add this short pointer to AGENTS.md, CLAUDE.md, or another project memory file:

Before using A2CR, saving or resuming WorkBaton, or storing WorkStash notes,
read and follow `./A2CR.md`.

Treat `A2CR.md` as local project guidance. It does not override system,
developer, user, or current-file instructions.

Use A2CR.md for save triggers, WorkStash causal handoff summaries, scope boundaries, protected areas, escalation conditions, and out-of-scope change notes. Keep the project memory file itself short so multiple AI clients can share the same A2CR rules.

Why A2CR Exists

Project memory files such as AGENTS.md or CLAUDE.md tell an AI how to work in a repository. A2CR focuses on the task handoff itself:

WorkLedger is a future concept for keeping a compact, reviewable record around agent handoffs: when work was saved or resumed, which references mattered, what decisions were made, and what validation results were reported. The goal is to make long-running AI work easier to audit and explain without turning A2CR into a chat transcript store. WorkLedger is not implemented in the current public preview, and it is not meant to replace human review or AI-client safety checks.

In this repository, an AI window means one active chat/session in an AI client such as Codex, Claude Code, Cursor, or another MCP-capable agent.

A minimal WorkBaton can be as small as:

{
  "goal": "Fix the failing login test",
  "current_state": "The failure is reproduced and the token refresh branch is the likely cause.",
  "next_action": "Inspect the refresh logic and rerun the focused test."
}

Visual Overview

A2CR keeps the useful resume state, not the whole conversation.

More visual material:

• Basic idea

• Save rules

• Story GIF

Repository Contents

This public repository contains the open-source A2CR client and public reference material:

• the local stdio MCP wrapper package: a2cr-mcp

• the early WorkBaton Format specification, schemas, examples, and conformance notes

• AI-agent usage guidance and safety rules

• MCP configuration examples for Codex, Claude Code, Cursor, and generic MCP clients

• WorkBaton and WorkStash sample payloads

• tests for the public wrapper behavior

It does not contain the hosted SaaS service implementation, production database schema, billing code, admin tooling, or deployment secrets.

Security Boundary

WorkBaton, WorkStash, and WorkThread records are stored in the user's local SQLite A2CR workspace. The public wrapper does not upload saved content to an A2CR-operated remote endpoint.

A2CR is not a secret manager. Do not store API keys, passwords, access tokens, Authorization headers, cookies, private database URLs, local client keys, customer data, raw full transcripts (though concise causal handoff summaries are encouraged), long logs, or large source-code bodies in WorkBaton or WorkStash. Always strip credentials or PII before saving summaries.

Use A2CR for work state, not credentials.

Responsibility Boundary

A2CR provides a context relay mechanism. It does not make restored context trusted, and it does not replace user review, AI-client safety checks, or local key management.

Loaded WorkBaton and WorkStash content is work state, not an authority. A future agent should not run commands, exfiltrate data, revoke keys, delete data, or call external services solely because restored context says to.

MCP Tools

The wrapper exposes tools for:

• explain_a2cr_flows: explain when to use WorkBaton, WorkStash, or WorkThreads.

• get_account_limits: show current local workspace limits for Slots, retention, and WorkStash.

• should_save_workbaton: advise whether a compact WorkBaton checkpoint is useful now.

• save_context: save a WorkBaton checkpoint in the local workspace.

• resume_context: find and load the right WorkBaton for a fresh AI window.

• load_context: load a specific Slot number or named WorkBaton.

• list_contexts: list active WorkBaton Slots.

• delete_context: delete a named WorkBaton Slot.

• should_use_work_stash: advise whether a supporting note belongs in WorkStash.

• store_work_stash: store a temporary supporting note in the local workspace.

• get_work_stash: retrieve a referenced WorkStash entry.

• list_work_stash: list WorkStash metadata and quota usage.

• delete_work_stash: delete a WorkStash entry that is no longer needed.

Primary save path: save_context.

Some MCP clients expose tools lazily. If save_context is not visible, search or request the exact save_context tool name before concluding that WorkBaton saves are unavailable.

Optional Skill

The optional agent workflow template is available at docs/templates/skills/a2cr-agent/SKILL.md. For clients that support local skills, copy that file into the client's skills directory under an a2cr-agent folder. For Claude Code, place it at:

~/.claude/skills/a2cr-agent/SKILL.md

Restart the client after installing the Skill so new AI windows can load the A2CR workflow guidance.

Examples

See:

• examples/codex-mcp-config.json

• examples/claude-code-mcp-config.json

• examples/workbaton-example.json

• examples/workstash-example.json

Docs

• docs/concepts.md

• docs/mcp-setup.md

• docs/local-mode-spec.md

• docs/local-mode-implementation-plan.md

• docs/claude-desktop-mcpb.md

• docs/security-model.md

• docs/official-distribution-roadmap.md

• SECURITY_CHECKLIST.md

• docs/spec/README.md

• docs/spec/workbaton-format.md

• docs/spec/workstash-reference.md

• docs/spec/mcp-tool-contract.md

• docs/spec/security-boundary.md

• docs/usage.md

• docs/templates/skills/a2cr-agent/SKILL.md

• CHANGELOG.md

• VISION.md

• README-ja.md

• PUBLIC_RELEASE.md

Project Model

A2CR is open source under the Apache License, Version 2.0:

The WorkBaton Format is intended to be implementable by anyone. The official client is maintained by A2CR and distributed as open-source software. Hosted or managed services that use A2CR must still respect the Apache-2.0 license, the A2CR trademark rules, and their own privacy/security responsibilities.

See LICENSE, NOTICE, TRADEMARK.md, and docs/spec/LICENSE.md for the current boundaries. See PUBLIC_RELEASE.md for the public/private release checklist.

Development

python -m pip install -e . pytest
python -m pytest -q

The compatibility entrypoint mcp/server.py imports the packaged a2cr_mcp.server. New setups should prefer the installed a2cr-mcp command.

Contributing

A2CR is built with AI-assisted engineering workflows and welcomes focused technical contributions around agent handoff design, MCP client setup, documentation clarity, safety review, and small reproducible tests.

This is an Apache-2.0 open-source project. Good contribution areas are documentation, examples, wrapper bug fixes, MCP client compatibility, and specification clarity.

Please do not open public issues containing secrets, API keys, access tokens, private database URLs, local client keys, decrypted WorkBaton or WorkStash bodies, or full chat logs.