DataMind-Plugin

Three packaging variants of the DataMind AI agent plugin — one for Codex, one for Claude Code, one for Cursor. They all expose the same 17 MCP tools (RAG, GraphRAG, Wiki, structured memory, feedback loop) over the same DataMind runtime, and they share the same on-disk data location so switching IDEs preserves your profiles and memories.

What does DataMind do?

It turns "chatting with an AI" into an accumulating local knowledge asset. Concretely:

• datamind_use_folder — point it at any folder of PDF/docx/xlsx/markdown/etc. and it builds a managed RAG + GraphRAG index, never touching your originals.
• datamind_ask — answer a question; the model gets Wiki snippets, structured memory, and retrieved chunks all merged in.
• datamind_remember — capture a durable preference, decision, or workflow.
• Wiki layer at ~/.datamind-context/profiles/{profile}/wiki/ — human-readable Markdown that accumulates as you ask high-value questions.
• Structured memory at ~/.datamind-context/memory.db — MemOS-style index of preferences/decisions/workflows.
• Feedback loop — every answer returns a feedback_trace.id; corrections become feedback_hint memory for similar future questions.

Full tool list is in any of the three sub-directories' USAGE.md.

Pick your IDE

DataMind-Plugin/
├── codex/datamind-context/         ← Codex users
├── claude-code/datamind-context/   ← Claude Code users
└── cursor/datamind-context/        ← Cursor users

Install in 3 lines

git clone https://github.com/OpenDCAI/DataMind-Plugin.git
cd DataMind-Plugin

# Pick the directory matching your IDE:
cd codex/datamind-context        && ./install.sh   # Codex
# cd claude-code/datamind-context && ./install.sh   # Claude Code
# cd cursor/datamind-context      && ./install.sh   # Cursor

Each install.sh is self-contained: it creates a Python venv under vendor/datamind/.venv, installs requirements.txt, registers the plugin with the corresponding IDE, and tells you where to put your LLM/embedding API keys.

After install, edit vendor/datamind/.env (each variant has its own copy under its install location) and restart your IDE.

Windows users

Windows isn't a second-class citizen. Each variant ships a PowerShell counterpart of every Bash script:

git clone https://github.com/OpenDCAI/DataMind-Plugin.git
cd DataMind-Plugin

# Pick one:
cd codex\datamind-context        ; .\install.ps1   # Codex
# cd claude-code\datamind-context ; .\install.ps1   # Claude Code
# cd cursor\datamind-context      ; .\install.ps1   # Cursor

install.ps1 does three things install.sh doesn't on Unix:

1. Creates the venv under vendor\datamind\.venv\Scripts\python.exe (Windows venv layout).
2. Replaces the bundled mcp.json with mcp.windows.json, so the IDE spawns powershell.exe -File run_datamind_mcp.ps1 instead of trying to execute the Bash launcher.
3. Honors -Force, -SkipDeps, -RepoRoot, -PythonExe flags (PowerShell-style equivalents of the Bash --force, --skip-deps, etc.).

If your PowerShell execution policy blocks the script, run it once with:

powershell -ExecutionPolicy Bypass -File .\install.ps1

Per-IDE differences

	Codex	Claude Code	Cursor
Manifest	.codex-plugin/plugin.json	.claude-plugin/plugin.json	.cursor-plugin/plugin.json
MCP config	.mcp.json (relative path)	.claude-plugin/mcp.json (uses ${CLAUDE_PLUGIN_ROOT})	mcp.json (uses ${userHome})
Install location	~/.codex/marketplaces/datamind/plugins/datamind-context/	Claude Code plugin cache (auto-managed)	~/.cursor/plugins/local/datamind-context/
Auto-trigger	Skills (SKILL.md)	Skills (SKILL.md)	Skills + a global Cursor Rule installed to ~/.cursor/rules/datamind.mdc
Marketplace	~/.codex/config.toml (auto-written)	claude plugin marketplace add	Auto-discovered from ~/.cursor/plugins/local/
Verify	Codex plugin list	claude mcp list should show ✓ Connected	Cursor Settings → Plugins → Local Plugins

Cross-IDE data sharing

All three variants read and write the same paths under ~/.datamind-context/. So if you build a profile in Codex and then switch to Claude Code, the same profile, Wiki, and memory are all there — no migration step needed.

~/.datamind-context/
├── profiles.json                  # Profile registry
├── profiles/{profile}/data/       # Managed copies of source files
├── profiles/{profile}/wiki/       # Per-profile Markdown Wiki
├── memory.db                      # Structured memory (SQLite + FTS)
├── interactions/{session}/        # Dynamic event logs
├── memories/{session}/            # Consolidated memories
└── skills/{session}/              # Generated skill markdown

Maintenance: the three copies

Each sub-directory is a complete, independent copy of the plugin — including src/*.py and vendor/datamind/. This lets you git clone and use any one of the three without depending on a build step or symlinks.

The trade-off: when you change the core MCP server or vendored DataMind code, you have to keep the three copies in sync. Use scripts/sync-core.sh (treats codex/datamind-context/ as the source of truth) to copy src/*.py and vendor/datamind/ to the other two.

./scripts/sync-core.sh             # Sync core code from codex/ to claude-code/ and cursor/
./scripts/sync-core.sh --check     # Dry-run; report differences without writing

Files that are not synced (each variant has its own):

• install.sh
• The IDE-specific manifest (.{codex,claude,cursor}-plugin/)
• The IDE-specific MCP config (.mcp.json, .claude-plugin/mcp.json, or mcp.json)
• README.md, INSTALL.md, USAGE.md, RELEASE_NOTES.md
• Cursor: .cursor/rules/datamind.mdc
• Claude Code: hooks/, scripts/bootstrap_claude.sh, .claude-plugin/marketplace.json
• The string "Codex" / "Claude Code" / "Cursor" inside skills/datamind-context/SKILL.md

When you tag a release, you can either:

1. Single repo + GitHub Release zips: zip each sub-directory separately as datamind-context-plugin-codex-X.Y.Z.zip etc.
2. Single repo, users git clone directly: tell users to cd <variant>/datamind-context && ./install.sh.

Both flows work.

Repo structure

DataMind-Plugin/
├── README.md                       # This file
├── .gitignore
├── scripts/
│   └── sync-core.sh                # Maintenance: keep three copies in sync
├── codex/
│   └── datamind-context/           # Self-contained Codex variant
├── claude-code/
│   └── datamind-context/           # Self-contained Claude Code variant
└── cursor/
    └── datamind-context/           # Self-contained Cursor variant

License

MIT, same as upstream DataMind.

Links

• DataMind upstream: https://github.com/zwt233/DataMind