The Veteran Resource Data Commons

1. Abstract

We describe the design and deployment of the first open-data, machine-queryable veteran-resource commons in the United States. The system indexes 20,000+ live-verified veteran-serving resources across all 50 states, the District of Columbia, Puerto Rico, Guam, the U.S. Virgin Islands, American Samoa, and the Northern Mariana Islands — exposed through a documented 13-endpoint REST API, a Model Context Protocol (MCP) server compatible with any MCP client, a Google A2A agent manifest, an llms.txt manifest at the domain root, and a SHA-256-integrity-hashed full dataset snapshot. The infrastructure operates at a unit cost of approximately $134 per verified resource per year, against a three-year program-expense ratio of 88%. The codebase is licensed MIT; the data is licensed CC BY 4.0. To our knowledge, no other U.S. veteran-serving 501(c)(3) publishes a comparable stack. This report documents the architecture, governance, and impact metrics, and discusses implications for grantmakers, civic-technology researchers, and peer nonprofits.

2. The problem: fragmentation of veteran services

The United States funds a vast veteran-services ecosystem. The U.S. Department of Veterans Affairs operates the nation's largest integrated health-care system, 317 hospitals, and hundreds of clinics. Every state runs a Department of Veterans Affairs or equivalent. Every county in most states operates a County Veterans Service Office (CVSO) or equivalent intake point. Veterans Service Organizations (VSOs) — American Legion, VFW, DAV, AMVETS — maintain thousands of post locations. A growing network of non-governmental direct-service providers covers transitional housing, crisis response, employment, and specialty care.

Yet a veteran searching for help at 2 a.m. routinely cannot find it. The VA reports that 22 veterans die by suicide every day [1]. HUD counts approximately 35,000 veterans unhoused on any given night [2]. The VA backlog of pending disability claims exceeds 260,000 [3]. The systemic cause is not absence of services — it is a navigation failure: stale phone numbers, relocated offices, obsolete eligibility rules, and fragmented data silos that do not interoperate.

Peer-sized veteran nonprofits address this failure largely through call centers and human navigators. That model serves those who call during business hours on a weekday; it does not scale to the crisis-window reality of veteran need. It also embeds the data in proprietary systems that are not auditable, not reproducible, and not machine-queryable by the AI assistants a growing share of veterans increasingly consult.

3. The intervention: an open-data commons

Wounded Warriors — a Texas 501(c)(3) public charity founded in April 2021 — built the Veteran Resource Data Commons as an alternative architecture: a single machine-queryable directory of every veteran-serving resource we can identify, verified by human staff on a rolling quarterly cycle, published under an open license, and exposed through every interoperability standard a modern AI assistant or research team can consume.

The commons exists simultaneously as:

• A consumer-facing search interface at warriorsfund.org, free, no login, crisis-line-routed by default;
• A 13-endpoint REST API documented in OpenAPI 3.0 [spec];
• A Model Context Protocol (MCP) server exposing five tools, listed on the Smithery registry [smithery];
• A Google A2A agent manifest at /.well-known/agent.json;
• An llms.txt manifest at the domain root for LLM discoverability;
• A SHA-256-integrity-verified dataset export at /api/export/snapshot;
• A grantmaker-grade organizational dossier endpoint at /api/grantmaker/snapshot, returning live impact, governance, financial, and trust-signal data.

Every endpoint is free. Every endpoint requires no authentication. The entire repository is discoverable through public registries (Google Dataset Search, Smithery, and the GitHub organization @WoundedWarriors).

4. Technical architecture

The system is deployed on Cloudflare's edge infrastructure for cost, latency, and uptime reasons.

4.1 The MCP server

The Model Context Protocol, introduced by Anthropic in November 2024, standardizes how AI agents connect to data and tools. Our server exposes five tools, each a JSON-RPC call against the production D1 database:

• search_veteran_resources(zip, type, radius, limit) — ZIP-based nearest-neighbor search with resource-type filter.
• get_veteran_resource_stats() — live aggregate statistics.
• veteran_crisis_resources(zip?) — 988 Veterans Crisis Line routing plus the nearest Vet Center. AI clients are instructed to call this unconditionally whenever user input contains crisis language.
• search_by_city(city, state?) — same as the ZIP variant but accepts natural place-names.
• calculate_va_benefits(type, rating?, serviceYears?) — deterministic estimator for VA disability compensation and GI Bill eligibility.

The MCP server is registered on Smithery, the canonical MCP server registry. Any MCP-compatible client — Claude Desktop, Cursor, Continue, and others — can install and use the server with a single command:

npx @smithery/cli install wounded-warriors-veterans --client claude

4.2 Crisis-routing as a first-class primitive

The API and the MCP server both embed a crisis-language detector with 25 keyword phrases covering explicit and indirect expressions of suicidal ideation, hopelessness, and emotional distress. When triggered on any inbound query, the system substitutes a response that surfaces:

1. The Veterans Crisis Line (988, press 1; text 838255; chat) as the primary action;
2. The geographically nearest Vet Center and mental-health resource from the database;
3. A non-clinical, de-escalation-framed acknowledgment message.

This behavior is documented in the MCP tool descriptions, so downstream AI agents are explicitly instructed to surface crisis resources rather than route a distressed veteran through a generic response path.

4.3 Data integrity and reproducibility

The /api/export/snapshot endpoint returns the complete current dataset with a SHA-256 hash computed over a canonical serialization. Any third party — a grantmaker, researcher, or peer organization — can download the snapshot, recompute the hash, and verify that the data served matches the data published. This mechanism satisfies the integrity requirement that many research institutional review boards and funder diligence teams impose without requiring on-chain or cryptographic infrastructure beyond standard hashing.

5. Impact & unit economics

All figures below are independently verifiable against the live API and against IRS Form 990 filings published on ProPublica Nonprofit Explorer.

Resource composition at the time of this report:

5.1 Unit-cost comparison with peer organizations

Larger veteran nonprofits in the same functional category — Wounded Warrior Project, Tunnel to Towers, Semper Fi & America's Fund, Gary Sinise Foundation, Team Rubicon — operate call-center or human-navigator models at materially higher unit cost per resource or per veteran contact. While published program-cost data varies by organization and methodology, internal cost-per-outcome estimates for those operating models are generally in the thousands of dollars per unique veteran contacted, not the hundreds of dollars per resource per year. The infrastructure-layer approach documented here is not a replacement for human support; it is an orthogonal efficiency wedge that multiplies the reach of every dollar those larger organizations raise.

6. Governance & trust signals

The organization is a Texas-domiciled 501(c)(3) public charity, ruling date April 2021, EIN 86-1336741. The Board of Directors has adopted and publishes five governance policies:

• Conflict of Interest Policy (modeled on IRS Form 1023 Appendix A)
• Whistleblower Policy (Sarbanes-Oxley §1107 compliant)
• Records Retention Policy
• Donor Privacy Policy (AFP Donor Bill of Rights, BBB WGA Standard 19)
• Gift Acceptance Policy

Third-party accountability ratings:

Independent audit: A full-scope audit under Generally Accepted Auditing Standards is engaged for FY2024 with a qualified Texas nonprofit-audit firm; report delivery is expected in Q3 2026. Prior fiscal years were unaudited, consistent with the Texas state threshold for mandatory audits.

7. Replicability for other nonprofits

The architecture described here is not specific to veteran services. Any nonprofit operating a directory of verified resources — food banks, housing programs, reentry services, disability advocacy, rural health — can adopt the same pattern:

1. Edge-deployed API (Cloudflare Workers, AWS Lambda@Edge, Vercel Functions): sub-$100/month at the Wounded Warriors scale, often free below metered thresholds.
2. OpenAPI 3.0 specification: the minimum bar for AI-agent discoverability. Auto-generatable from most modern web frameworks.
3. MCP server exposing the core read operations: listable from the Smithery registry, installable into Claude Desktop and Cursor with one command.
4. llms.txt manifest at the domain root: emerging standard for LLM-oriented site metadata; trivial to produce.
5. SHA-256-hashed full dataset export: single endpoint, hash computed at request time, no additional infrastructure.
6. Crisis-routing primitive: a keyword detector and a structured crisis response. Where sector-appropriate (suicide prevention, domestic violence, substance-use disorders), this should be a first-class API capability rather than a feature flag.
7. Governance policies: published publicly. Board-adopted. Reviewed annually. Free templates from Council of Nonprofits and from this project's own repository.
8. Data license: CC BY 4.0 or a compatible open license. Without this, downstream research and AI-agent consumption are legally ambiguous.

The full source code for the MCP server, the API, and the governance-policy templates is available at github.com/WoundedWarriors/wounded-warriors-mcp. The data is at warriors-fund-api.emperormew.workers.dev/api/export/snapshot.

8. Implications for researchers

The dataset is, to our knowledge, the largest publicly available structured directory of U.S. veteran-serving resources with associated geocoding, verification timestamps, and crisis-routing annotations. It is suitable for:

• Access-to-care studies. The 56-jurisdiction coverage and ZIP-level geocoding enable distance-to-nearest-service analyses across rural/urban gradients, territorial vs. mainland jurisdictions, and county-level coverage gradients.
• AI-agent benchmarking. The MCP server and structured tool surface provide a non-adversarial, bounded-risk environment for evaluating how agents handle crisis language, resource retrieval, and benefit-calculation queries.
• Service-ecosystem mapping. Cross-referenced with VA operations data, HUD AHAR figures, and Census veteran-population tables, the dataset can support policy analyses of service density relative to veteran need.
• Replication and longitudinal study. Quarterly snapshots with SHA-256 attestation allow researchers to track changes over time without operational dependencies on the host organization.

Researchers wishing to partner on published work are encouraged to contact info@warriorsfund.org. Citations using the format in the header of this document are welcomed.

9. Future directions

Several extensions are under consideration or active development:

1. Server-sent events feed at /api/grantmaker/live-feed: a real-time stream of resource additions, verifications, and anonymized crisis-router events, suitable for impact dashboards and grantmaker observability.
2. Federated verification partnerships: inviting peer veteran nonprofits to submit resources they verify, with attribution preserved and provenance auditable.
3. Academic-partner research collaborations, especially with institutions conducting veteran-health and service-access studies.
4. Corporate matching-platform integrations (Benevity, YourCause, Bright Funds), leveraging the grantmaker-snapshot endpoint as the canonical source of impact data for employee giving campaigns.
5. Expanded crisis-router coverage: from 25 to 50+ phrase categories, calibrated against VA public-health data on help-seeking language.
6. Increased data density in under-served geographies: particularly Alaska, Guam, American Samoa, and the Northern Mariana Islands, where current resource counts are below median.

10. Data access & licensing

11. References

1. U.S. Department of Veterans Affairs. National Veteran Suicide Prevention Annual Report, 2023. mentalhealth.va.gov/suicide_prevention/data.asp
2. U.S. Department of Housing and Urban Development. Annual Homeless Assessment Report (AHAR) to Congress, 2023.
3. U.S. Department of Veterans Affairs. Monday Morning Workload Report (weekly).
4. Anthropic. Model Context Protocol Specification, 2024. modelcontextprotocol.io
5. Smithery. MCP server registry. smithery.ai
6. ProPublica Nonprofit Explorer, EIN 86-1336741. projects.propublica.org/nonprofits/organizations/861336741
7. Candid (GuideStar). Seal of Transparency program. candid.org
8. Charity Navigator. Encompass Rating System. charitynavigator.org/about-us/our-methodology/ratings
9. Creative Commons. Attribution 4.0 International License. creativecommons.org/licenses/by/4.0
10. Association of Fundraising Professionals et al. Donor Bill of Rights. afpglobal.org/donor-bill-rights