bncc-pacotes

The bncc-pacotes repository provides machine interfaces for the Base Nacional Comum Curricular (BNCC), the official dataset of Brazilian basic education. It contains three publications derived from the [bncc-dados](https://github.com/bncc-dev/bncc-dados) dataset: an npm package (`@bncc/dados`) for TypeScript/JavaScript, a PyPI package (`bncc`) for Python, and an MCP server (`@bncc/mcp`) for integration with AI agents. The data (1,721 learnings) is embedded directly into the packages, eliminating runtime dependencies and the need for network access. Queries run locally on the user's machine. The dataset is synchronized from a specific commit of the bncc-dados repository, with checksums and a data-version recorded in `dados/VERSAO.json`. The API offers functions such as `porCodigo`/`por_codigo` for lookup by code (e.g., `EF67LP08`), `habilidadesEF`/`habilidades_ef` to filter by component, year, and practice, `buscar`/`buscar` for text search, and `decodificar`/`decodificar` to decompose codes into their constitutive parts (stage, area, competency, sequence). The MCP server exposes seven tools (`bncc_lookup`, `bncc_buscar`, `bncc_listar`, `bncc_decodificar`, `bncc_estatisticas`, `bncc_estrutura`, `bncc_progressao_ei`) that allow AI agents to query BNCC with the official source cited in each response. Nonexistent codes return explanatory errors about legitimate gaps in the numbering, avoiding hallucinations. The server can be used locally via `npx -y @bncc/mcp` or accessed freely at `https://mcp.bncc.dev` via HTTP streamable, without an API key. Parity between the npm and PyPI implementations is ensured by "golden" queries tested on every commit, as documented in [docs/paridade.md](docs/paridade.md). The project is maintained by [Profy](https://www.profy.ai/), a company that operates LLM-based products over the BNCC. The code is licensed under MIT and the data under CC BY 4.0, in separate license files. The "bncc.dev" name and visual identity are not licensed.