MDD memory from the command line, an MCP server and the VS Code extension

The engineering memory of each project (see the MDD update) only works if it is easy to maintain. This round, built in the repository that distributes the agents' skills and rules, delivers the tools for that.

The mdd client

One Python command for any project in the ecosystem:

• mdd init creates the memory/ folder with the three base files and the indexes.
• mdd sync syncs the canonical skills, compared by SHA-256.
• mdd compact applies the ten-active-items rule and the size ceiling, keeping both the original and the summarized version.
• mdd status and mdd report show the health of the memory and export the report.
• mdd index and mdd meta set/get rebuild the indexes from each file's header. The core does the same in PHP (c2f memory:index, memory:set, memory:get), and 48 test cases check that both versions give the same result.

Hub and documentation watcher

The hub receives project reports through a local API, with a token, and refuses duplicate or malformed submissions. The watcher follows the official documentation of the agent tools and records what changed, never touching anything on its own outside a dedicated branch.

MCP server and extension

mdd mcp exposes nine tools to any compatible agent, over stdio or SSE on the local address. The VS Code extension (version 1.1.3) shows the MDD memory, looks for memory/ before sdd/, and opens a skill-conflict panel: when a local copy of a skill differs from the source, you can update it or mark the difference as intentional.

Status

Client and hub tests (125, no failures), extension tests (131) and real calls to the MCP server. The independent review recommended approval; the visual acceptance of the extension is still up to the Chief Engineer, and the extension package has not been published.

Source: AI Workspace · REQ-069, REQ-072 and REQ-074 · BATCH-071, BATCH-074 and BATCH-076