SAM.gov + USAspending + Federal Register + eCFR + Grants.gov skill

This skill teaches Claude how to use the `sam-gov` MCP server (36 keyless tools wrapped from @cliwant/mcp-sam-gov) to answer the full surface of US federal contracting / spending / regulation questions end-to-end.

Setup requirement: the sam-gov MCP server must be reachable. If you installed via /plugin install seungdo-keum/mcp-sam-gov, the bundled .mcp.json registers the server automatically. Otherwise see the repo README for manual MCP setup.

Available tools — 36 total, all keyless

The MCP server exposes the only sources of truth for federal contracting + spending + regulation data — never invent notice IDs, contracting officer names, award amounts, NAICS codes, or regulation citations.

SAM.gov — opportunities + attachments (5 tools)

USAspending — awards + recipients (8 tools)

USAspending — aggregate analysis (6 tools)

USAspending — agency profile (3 tools)

USAspending — recipient profile (2 tools)

USAspending — reference / autocomplete (5 tools)

Federal Register — rules + notices (3 tools)

eCFR — Code of Federal Regulations (2 tools)

Grants.gov — federal grants (2 tools)

The exact tool names depend on the host — Claude Code prefixes MCP tools as mcp__<server>__<tool>, while bare MCP hosts use just <tool>. Use whichever your host gives you.

7 standard workflows

Workflow 1 — Discover + qualify a single opportunity

1. sam_search_opportunities with NAICS / agency / state filter.
2. For most promising hit: sam_get_opportunity with its noticeId.
3. SOW depth: sam_fetch_description for full RFP text.
4. Attachments: surface URLs from step 2's attachments array.
5. (Optional) Cross-check the issuing agency with usas_get_agency_profile for context.

Workflow 2 — Competitive landscape ("who wins at agency X")

1. If agency is an abbreviation: usas_lookup_agency first to get canonical name.
2. usas_search_awards with canonical name + NAICS + fiscal year.
3. Line items: usas_search_individual_awards.
4. Office-level: usas_search_subagency_spending.
5. (Optional) Time-series: usas_spending_over_time to see trend.

Workflow 3 — Recompete radar

1. usas_search_expiring_contracts for agency × NAICS × N months.
2. For each candidate: usas_get_award_detail to get period_of_performance + set-aside + competition extent.
3. Cross-reference with sam_search_opportunities (same NAICS) to find pre-RFP shaping (Sources Sought / Pre-solicitation notices).

Workflow 4 — Teaming / supply-chain map

1. usas_search_individual_awards for prime awards at target agency.
2. usas_search_subawards filtered by prime to surface sub network.
3. For each sub: usas_get_recipient_profile to confirm size standards / alternate names.

Workflow 5 — Capture brief / agency intelligence (NEW)

1. usas_lookup_agency → canonical name + toptier code.
2. usas_get_agency_profile — mission + scale.
3. usas_get_agency_budget_function — where the budget actually goes.
4. usas_search_subagency_spending — buying offices.
5. usas_spending_over_time (group=fiscal_year) — multi-year trend.
6. usas_search_state_spending — geographic spend distribution.

Workflow 6 — Regulatory context for a pursuit (NEW)

When the user asks about FAR / DFARS / set-aside policy / cybersecurity rules:

1. ecfr_search (titleNumber=48 for FAR) — find the relevant section text.
2. fed_register_search_documents (agency + recent date range) — find proposed rules / amendments.
3. fed_register_get_document — pull the exact citation + body.
4. Quote at most 1 short snippet (< 15 words) from regulation text. Always include the eCFR section path or Federal Register citation.

Workflow 7 — Grants pivot (NEW)

When the user pivots from contracts to grants (very common for SDVOSB / 8(a)):

1. grants_search with keyword / CFDA.
2. grants_get_opportunity — full grant detail.
3. (Optional) usas_search_cfda_spending — see who's already winning that grant program.

Output discipline (anti-hallucination)

• Cite tool calls inline when surfacing facts: "VA awarded $410M to Booz Allen across 28 contracts in FY26 (usas_search_awards)."
• Never invent notice IDs, NAICS codes, recipient names, contract amounts, regulation citations. If a tool returns nothing, say so.
• Use autocomplete guards FIRST:
• User mentions a NAICS theme without code → usas_autocomplete_naics to confirm.
• User mentions a recipient name → usas_autocomplete_recipient to confirm exact USAspending-canonical legal name.
• User mentions an agency abbreviation → usas_lookup_agency to get canonical name + toptier code.
• Quote sparingly from RFP body / regulation text — at most 1 snippet, < 15 words. Point to the source URL.
• Notice IDs starting with demo- are fictional fixtures from non-OSS demos; they don't exist on sam.gov.

Common pitfalls

• ❌ "VA" passed directly as agency to USAspending tools → returns nothing. ✅ usas_lookup_agency("VA") first.
• ❌ Searching opportunities without a ncode or query filter — useless deluge. ✅ Always narrow.
• ❌ Quoting an old noticeId from training data — they expire. ✅ Always do a fresh sam_search_opportunities.
• ❌ Calling sam_fetch_description before sam_get_opportunity — you don't have the noticeId yet.
• ❌ Confusing usas_search_cfda_spending (grants) with usas_search_psc_spending (contracts) — different award types.
• ❌ Using a NAICS code from training without verification — they get retired. ✅ usas_autocomplete_naics to confirm currency.
• ❌ Agency name spelled wrong (e.g. "VA Department" vs canonical "Department of Veterans Affairs") → empty results. ✅ Always use usas_lookup_agency.

Optional — higher rate limits + archives

The MCP server runs keyless by default. For higher SAM.gov rate limits + archives older than ~12 months, the operator can set SAM_GOV_API_KEY in the MCP server's env block. The agent doesn't need to know — auth path is transparent.