Open Census MCP Server
# Open Census MCP Server
[](#disclaimer)
## Disclaimer <a name="disclaimer"></a>
This is an **independent, open-source experiment**.
It is **not** affiliated with, endorsed by, or sponsored by the U.S. Census Bureau or the Department of Commerce.
Data retrieved through this project remains subject to the terms of the original data providers (e.g., Census API Terms of Service).
## What Is This?
An AI-powered statistical consultant for U.S. Census data. Ask questions in plain English, get accurate demographic data with proper statistical context, methodology guidance, and fitness-for-use caveats.
**The insight:** Census data has a pragmatics problem, not a search problem. Knowing WHICH data to use and HOW to interpret it matters more than finding it. This system encodes statistical consulting expertise into the AI interaction layer.
## Status
š¬ **Active Research & Rebuild** ā v3 architecture in progress.
See [docs/lessons_learned/](docs/lessons_learned/) for the v1/v2 journey.
## Vision
Census data influences billions in policy decisions, but accessing it effectively requires specialized knowledge. This project aims to make America's most valuable public dataset as easy to use as asking a question ā with the statistical rigor of a professional consultant.
**The opportunity:** Every city council member, journalist, nonprofit director, and curious citizen should be able to fact-check claims and understand their communities with the same ease an eighth-grader uses a search engine. The data is public. The expertise to use it properly shouldn't be gatekept by technical complexity.
## Architecture (v3)
Pure Python MCP server with pragmatic rules engine. No R dependency.
- **Pragmatic Rules Layer:** Fitness-for-use constraints (MOE thresholds, coverage bias, temporal validity, source selection)
- **Census API Integration:** Direct Python calls to Census Bureau APIs
- **Knowledge Base:** Methodology documentation for RAG-enhanced guidance
Details: [docs/architecture/](docs/architecture/) (coming soon)
## Project Structure
```
docs/ # Systems engineering documentation
requirements/ # ConOps, SRS
architecture/ # System architecture
decisions/ # ADRs, trade studies
design/ # Detailed design
verification/ # V&V, evaluation results
lessons_learned/ # Project narrative & lessons
knowledge-base/ # Source docs & pragmatic rules
source-docs/ # Census methodology PDFs (gitignored)
rules/ # Extracted pragmatic rules
methodology/ # Processed methodology content
src/ # MCP server source code
tests/ # Evaluation harness & unit tests
scripts/ # Build & utility scripts
```
## Acknowledgments
- **U.S. Census Bureau** ā for collecting and maintaining vital public data
- **Kyle Walker** ā [Analyzing US Census Data](https://walker-data.com/census-r/) textbook as knowledge base source
- **Anthropic** ā Model Context Protocol enabling AI tool integration
## Contributing
Contributions welcome, especially:
- Domain expertise from Census data veterans
- Statistical methodology review
- Evaluation test cases (real-world query scenarios)
## License
MIT License - see [LICENSE](LICENSE) file for details.
TDQS
Scored across 3 tools
Each tool has a clearly distinct purpose with no overlap: explore_variables for variable discovery, get_methodology_guidance for statistical methodology orientation, and get_census_data for data retrieval. The descriptions explicitly guide when to use each tool, eliminating any ambiguity.
All three tools follow a consistent verb_noun pattern with underscores: explore_variables, get_methodology_guidance, and get_census_data. The naming is uniform and predictable across the set.
With only 3 tools, the set feels thin for a Census data domain that typically involves complex workflows like filtering, aggregation, or visualization. While the tools cover core functions, the low count may limit agent capabilities for comprehensive analysis.
The tools cover key aspects: methodology guidance, variable discovery, and data retrieval, forming a logical workflow. However, there are minor gaps such as no tools for data transformation, geographic mapping, or advanced filtering, which agents might need to work around.