@policyrisk/mcp-server

PRIA MCP server — the policy risk and assessment engine for AI agents. Gives Claude, Gemini, and other MCP-aware models direct access to U.S. statutes (USC), federal regulations (CFR), pending bills, the Federal Register, the Polipedia knowledge wiki, and a policy knowledge graph — plus a personalized risk score for the calling user.

A policy-risk engine for AI agents. One MCP server, the entire U.S. policy stack.

Drop this into Claude, Cursor, Gemini, or any MCP-aware client and your model gets first-class access to:

• U.S. Code and Code of Federal Regulations — full text, search, citations
• Federal bills and the Federal Register — pending legislation, final rules
• Polipedia — a search-first wiki covering every U.S. policy topic
• Policy knowledge graph — traverse statutes ↔ bills ↔ rules ↔ executive orders
• PRIA Score — a personalized policy-risk score for the calling user
• Multi-agent research and deliberation — synthesized answers with permalinks
• 11 deterministic tax calculators — SALT, IRMAA, Roth conversion, QBI, …

Every response includes a permalink so the model can cite a page the user can actually open.

Install

Pick your client. Each one points at npx @policyrisk/mcp-server, which fetches the latest release on first run.

Claude Code

claude mcp add pria -- npx @policyrisk/mcp-server

Then export PRIA_API_KEY=pria_… (or paste it into the entry Claude Code wrote to ~/.claude.json).

Claude Desktop · Cursor · VS Code · Gemini CLI

Add this block to your client's MCP config (claude_desktop_config.json, .cursor/mcp.json, .vscode/mcp.json, ~/.gemini/settings.json, …):

{
  "mcpServers": {
    "pria": {
      "command": "npx",
      "args": ["@policyrisk/mcp-server"],
      "env": { "PRIA_API_KEY": "pria_…" }
    }
  }
}

Restart the client. The Polipedia wiki tools work without a key; everything else needs PRIA_API_KEY.

Get an API key

Two paths — pick whichever fits the agent's flow.

Web — log in at policyrisk.com, open Settings → API Keys, generate a key. Best when a human is in the loop.

Agent signup — no web account needed:

curl -X POST https://api.policyrisk.com/api/v1/signup \
  -H 'content-type: application/json' \
  -d '{"email":"[email protected]","firstName":"Ada","lastName":"Lovelace","dob":"1990-01-01"}'

Returns { apiKey: "pria_…" } exactly once. Store it; it isn't shown again.

Tools

A consolidated surface (snake_case for Gemini compatibility) plus the legacy hyphenated tool names for clients pinned to earlier versions. Each consolidated tool has a discriminator (mode, view, action, source, kind, or calculator) that selects the specific capability.

Personalized + policy data — requires PRIA_API_KEY

| Tool | What it does | |---|---| | pria_score | The user's PRIA Score. mode: 'summary' \| 'detailed' \| 'stress-test' \| 'export-pdf'. | | policy_exposures | Risks hitting the household. view: 'list' \| 'dollars' \| 'actions' \| 'budget-drilldown'. | | alerts | Custom watch rules + their firings. action: 'list-fired' \| 'list-rules' \| 'create' \| 'update' \| 'delete'. | | pins | Dashboard pin management. action: 'list' \| 'propose' \| 'remove'. | | api_keys | Self-service key management. action: 'list' \| 'create' \| 'revoke'. | | research | Multi-agent synthesized research with citations + permalinks. Fans out to bills, FR, policy intel, tariffs, and Polipedia. mode: 'sync' \| 'async-start' \| 'async-result', kind: 'question' \| 'fact-check-headline'. | | deliberation | Multi-LLM deliberation (panel + debate + scoring + judge) for high-stakes questions. mode: 'start' \| 'result'. | | briefing | Curated feeds. kind: 'daily' \| 'risk-events'. | | assess_policy | Submit arbitrary policy text → similar enacted statutes, pending bills, FR rules, risk signals. | | tax_calc | 11 profile-aware deterministic tax calculators. calculator: 'salt' \| 'section-179' \| 'irmaa' \| 'roth-conv' \| 'backdoor-roth' \| 'marriage-penalty' \| 'aca-subsidy' \| 'qbi' \| 'ss-breakeven' \| 'hsa' \| 'estate-stepup'. |

Source-spanning — wiki source needs no key

| Tool | What it does | |---|---| | lookup | Fetch one document by id. source: 'usc' \| 'cfr' \| 'federal-register' \| 'bill' \| 'wiki' \| 'fred-series' \| 'hts-tariff'. | | search | Search the policy corpus or traverse the knowledge graph. source: 'usc' \| 'cfr' \| 'federal-register' \| 'bills' \| 'policy-intel' \| 'wiki' \| 'related-graph'. |

Try it

Once the server is wired up, ask the model things like:

| Prompt | Tool call | |---|---| | "What's my PRIA Score?" | pria_score(mode: 'summary') | | "How does the new tax bill affect me?" | research(mode: 'sync', query: '…') | | "Pull 26 USC § 32 and explain the EITC phase-out." | lookup(source: 'usc', title: 26, section: '32') | | "Find every CFR rule citing 42 USC 1396a." | search(source: 'related-graph', nodeType: 'USC_SECTION', sourceId: '…') | | "Here's a draft executive order — what existing policies overlap?" | assess_policy(policyText: '…') | | "What's happened in policy today?" | briefing(kind: 'daily') | | "Search Polipedia for SALT cap changes." | search(source: 'wiki', query: 'SALT cap') |

Behind the scenes the model picks tools, chains calls, and cites permalinks back to policyrisk.com so the user can audit the answer.

Migration: 0.3.x → 0.4.x

The 0.3.x surface had 60 separately-named hyphenated tools. 0.4.x adds a consolidated 12-tool surface with discriminator parameters; the legacy hyphenated names remain available for clients pinned to earlier versions.

| 0.3.x tool | 0.4.x call | |---|---| | get-pria-score | pria_score(mode: 'summary') | | get-pria-score-detailed | pria_score(mode: 'detailed') | | run-monte-carlo-stress-test | pria_score(mode: 'stress-test') | | export-pria-scan-pdf | pria_score(mode: 'export-pdf') | | get-policy-exposures | policy_exposures(view: 'list') | | get-policy-exposure-dollars | policy_exposures(view: 'dollars') | | get-action-items / get-recommended-actions-detailed | policy_exposures(view: 'actions') | | get-budget-impact / get-budget-impact-detailed | policy_exposures(view: 'budget-drilldown') | | get-alerts | alerts(action: 'list-fired') | | list-alert-rules / create-alert-rule / update-alert-rule / delete-alert-rule | alerts(action: 'list-rules' \| 'create' \| 'update' \| 'delete') | | propose-alert-from-conversation | removed — fill the structured draft yourself, call alerts(action: 'create') | | list-pins / propose-pin / unpin | pins(action: 'list' \| 'propose' \| 'remove') | | list-my-keys / create-key / revoke-key | api_keys(action: 'list' \| 'create' \| 'revoke') | | lookup-usc-section / lookup-cfr-section / lookup-federal-register-rule / get-bill / get-polipedia-page / get-economic-indicator / get-tariff-impact | lookup(source: …) | | search-usc / search-cfr / search-federal-register / search-legislation / search-policy-intel / search-polipedia / get-related-policies | search(source: …) | | research-policy / start-research-async / get-research-status / get-research-result | research(mode: 'sync' \| 'async-start' \| 'async-result') (async-result returns 202 if pending — no separate status poll) | | fact-check-headline | research(mode: 'sync', kind: 'fact-check-headline') | | ask-polipedia | Still available as a legacy tool, or call research(mode: 'sync') for fanout. | | start-deliberation / get-deliberation-status / get-deliberation-result | deliberation(mode: 'start' \| 'result') | | get-daily-briefing / get-risk-events | briefing(kind: 'daily' \| 'risk-events') | | assess-policy-risk | assess_policy | | calc-* (11 calculators) | tax_calc(calculator: …) |

Requirements

• Node.js 18+
• PRIA_API_KEY (free — email + name + DOB)
• An MCP-aware client (Claude Desktop, Claude Code, Cursor, VS Code, Gemini CLI, …) with its own model API key

Links

• npm — https://www.npmjs.com/package/@policyrisk/mcp-server
• Source — https://github.com/PRIA-Technologies/skunkworks/tree/main/packages/mcp-server
• Web app — https://policyrisk.com
• Client compatibility notes — COMPAT.md

License

MIT