Code for America and Anthropic have partnered to develop AI tools for government caseworkers, starting with a Claude-powered integration designed to streamline the administration of public benefits programs like the Supplemental Nutrition Assistance Program (SNAP). The collaboration focuses on reducing bureaucratic complexity and improving efficiency for caseworkers navigating frequent policy changes.
Overview
The partnership introduces the SNAP Policy Navigator, a Claude-based tool that provides caseworkers with real-time, verified answers to policy questions. Built on Anthropic’s Model Context Protocol (MCP), the system ensures responses are grounded in up-to-date federal, state, and county regulations. MCP, an open standard adopted across the AI industry, enables secure connections between trusted data sources and AI applications, making it suitable for high-stakes environments like benefits administration.
The tool is part of a broader effort to modernize government service delivery. Code for America and Anthropic plan to expand the suite of integrations to include features like eligibility document review and plain-language communication drafting for benefit recipients. The goal is to create reusable, adaptable tools that can be deployed across states and counties.
What the tools do
The initial SNAP Policy Navigator addresses three core challenges:
- Policy navigation: Caseworkers can query complex, evolving SNAP rules and receive accurate, context-specific answers.
- Efficiency: The tool reduces manual research time, allowing faster case processing.
- Compliance: Responses are sourced from verified policy documents, minimizing errors.
Future integrations may include:
- Eligibility document review: Automated analysis of submitted documents to flag potential issues or missing information.
- Plain-language communications: Drafting clear, accessible messages for benefit recipients to explain decisions or next steps.
- Cross-program support: Extending the tools to other benefits programs beyond SNAP.
How it works
The SNAP Policy Navigator leverages retrieval-augmented generation (RAG) to pull real-time policy data from federal, state, and county sources. Caseworkers interact with the tool via a secure interface, posing questions in natural language (e.g., “What are the income limits for a household of four in California under the 2026 SNAP guidelines?”). The system retrieves the relevant policy snippets and generates a concise, cited response.
Anthropic’s Claude model powers the tool, with safeguards to ensure outputs are interpretable and steerable. The use of MCP ensures data privacy and compliance with government security standards.
Tradeoffs and limitations
- Data freshness: The tool’s accuracy depends on the timeliness of policy databases. Outdated or incomplete sources could lead to incorrect responses.
- Adoption barriers: Government agencies may face internal resistance or technical hurdles in integrating AI tools.
- Scope: The initial focus is on SNAP, leaving other benefits programs (e.g., Medicaid, TANF) unaddressed for now.
- Human oversight: While the tool aims to reduce errors, caseworkers must still review outputs for edge cases or ambiguous policies.
When to use it
The SNAP Policy Navigator is designed for:
- State and county agencies administering SNAP or other benefits programs.
- Caseworkers handling high caseloads with frequent policy changes.
- Governments seeking to modernize service delivery while maintaining compliance.
The partnership’s tools are not yet publicly available but are being piloted in select states. Agencies interested in adopting the technology can contact Code for America for deployment details.
Bottom line
The collaboration between Code for America and Anthropic represents a practical application of AI in government, targeting a critical pain point: bureaucratic inefficiency in benefits administration. By grounding responses in verified policy data, the tools aim to reduce errors, save time, and improve outcomes for both caseworkers and recipients. While challenges like data freshness and adoption remain, the partnership’s focus on reusable, scalable solutions could set a precedent for AI in public services.