zihangzzz-polaris-mcp 0.2.3

Local MCP memory layer for AI coding agents

Polaris

Polaris stops your AI coding agent from repeating the same mistakes. Open source. Pay $2.49 once if it saves you a loop.

Install

git clone https://github.com/zhaozihangbehappy-gif/Polaris.git && cd Polaris && python3 -m pip install -r adapters/mcp-polaris/requirements.txt

Optional CLI install from PyPI:

pipx install zihangzzz-polaris-mcp

Pay

https://<YOUR_HANDLE>.gumroad.com/l/polaris

One payment. Forever. The receipt email is your proof of support. No account, no login.

Contribute

Polaris grows through the people who use it. If Polaris saved you a loop — or if you hit a failure it didn't know about — you can add it:

• python3 scripts/polaris_community.py submit <your-pattern.json> — propose a new candidate. Accepts either a full v4 pattern record or a schema_version: 1 contribution file exported by polaris_cli.py experience contribute (auto-converted).
• python3 scripts/polaris_community.py confirm <pattern_id> — tell the project a candidate actually helped you on a real case.
• python3 scripts/polaris_community.py reject <pattern_id> — tell the project a candidate was wrong or harmful.

A candidate joins Polaris's shared library after ≥ 2 independent users confirm it helped them and no one rejected it. The shared library has three tiers — official (internal verified-live), community (community-verified through this channel), and candidate (unconfirmed but shipped so users can try them). All three are loaded at lookup time; each match carries its tier. Nobody's agent runs through a central server; confirmations are opt-in. See community/README.md for how trust and promotion work.

The shipped candidate tier now covers all 8 ecosystems. The official tier remains the audited baseline, while candidate breadth grows faster through releases and community submissions.

Read these, not the rest

• START_HERE.md
• FACTS.md
• INSTALL.md

There's an archive/ folder too. It's old audit work I kept public so anyone can check it. Don't read it to learn how Polaris works — the three files above are the whole product.

There's also an eval/ folder with historical evaluation fixtures and logs. Public, useful for auditing, not required for install or everyday use.

License

MIT