A2cr

What a2cr/a2cr does

The a2cr/a2cr MCP server provides a local workspace for handing coding tasks from one AI session to another. It separates the compact state needed to resume work from optional supporting notes. A WorkBaton can capture the goal, current state, decisions, blockers, validation results, and next action. WorkStash holds temporary details that would make the main checkpoint too large, with references connecting the two.

The workspace is intended for milestone checkpoints rather than full chat transcripts, secrets, large files, or a durable knowledge base. The material also describes WorkThreads as a feature in development and WorkLedger as a future concept, not as current public-preview functionality.

How it works

The a2cr/a2cr MCP server runs locally and stores WorkBaton, WorkStash, WorkThread, actor, and event records in SQLite on the user's machine. The public local wrapper does not use an A2CR account, API key, hosted relay, dashboard, or cloud sync path. Client-side encryption is part of the handoff model described for saved checkpoints and notes.

Use save_context to create a checkpoint and resume_context to locate the appropriate checkpoint when opening a fresh AI window. load_context retrieves a specific Slot or named WorkBaton, while list_contexts and delete_context manage saved Slots. WorkStash entries can be stored, retrieved, listed, and deleted separately. The advisory tools can help decide when a checkpoint or supporting note is appropriate.

A typical handoff keeps the WorkBaton short and places larger causal or supporting details in WorkStash. The next session can retrieve a referenced note using its entry key. The workspace exposes local limits for Slots, retention, and WorkStash through get_account_limits.

Setup and configuration

Install the Python distribution from PyPI with python -m pip install --upgrade a2cr-mcp. The compatibility executable a2cr-mcp starts the local workspace server for generic MCP clients. The README also identifies a2cr-local-mcp as the command used by the Codex registration flow.

A generic MCP configuration can launch a2cr-mcp with no arguments. Set A2CR_LOCAL_DB when an alternate SQLite database path is needed; the example path in the documentation is optional. Python 3.12 or 3.13 is recommended, while Python 3.15 development builds are not supported.

The project also documents a Node MCPB or Claude Desktop Extension distribution based on a GitHub Release asset. End users do not install that path through npm.

Tools and capabilities

• Explain when to use WorkBaton, WorkStash, or WorkThreads.
• Check local limits for Slots, retention, and WorkStash.
• Decide whether to save a WorkBaton or store a WorkStash note.
• Save, resume, load, list, and delete WorkBaton context.
• Store, retrieve, list, and delete WorkStash entries.

Limitations and notes

The a2cr/a2cr MCP server is designed for local handoffs, not cloud synchronization or hosted relay workflows. WorkStash is temporary supporting storage, not a durable knowledge base. WorkBaton is not intended for full transcripts, secrets, or large files. WorkLedger is not implemented in the current public preview and does not replace human review or client safety checks.

For project-specific behavior, the repository recommends an A2CR.md file in the project root. That file can define save triggers, scope boundaries, protected areas, escalation conditions, and out-of-scope changes. AI clients should read those local rules before saving or resuming context.

Getting started with this a2cr/a2cr MCP server

Always refer to the official documentation for the most accurate and up-to-date information.