SAM.gov & SBIR Federal RFP + Attachment Deep-Parser

Stop opening 40-page PDFs to find out you can't bid. This Actor finds new federal opportunities on SAM.gov and SBIR/STTR sources, downloads the attached RFP, SOW and PWS files, reads them for you, and returns the facts that decide a bid:

• Clearance level (Public Trust up to TS/SCI with Poly) and DD254
• Evaluation method (LPTA, Best Value Tradeoff or HTRO)
• Incumbent contract numbers (PIIDs) and incumbent contractor names
• Page limits and formatting rules for each volume
• Compliance frameworks (CMMC, NIST 800-171, FedRAMP, ITAR, IL5 and more)
• Every .gov and .mil email and phone number buried in the attachments

You don't need a SAM.gov API key or any setup. Press Start and the default search returns AI, sensor fusion and autonomy opportunities from the last 7 days.

---

Why this Actor exists

Most SAM.gov scrapers only return what the search page shows: title, solicitation number and posted date. The information that decides whether you can bid is in the attachments: the Statement of Work, the Section L instructions, the Section M evaluation criteria and the DD254. Capture teams lose hours every week opening those files one at a time.

This Actor opens them for you:

	Typical SAM.gov scraper	GovCon RFP Deep-Parser
Title, solicitation #, dates, agency	Yes	Yes
NAICS, PSC, set-aside, place of performance, POCs	Sometimes	Yes
Downloads PDF / DOCX / ZIP attachments	No	Yes, streamed in memory
Clearance, DD254, polygraph	No	Yes, with evidence quotes
LPTA vs. best value vs. HTRO	No	Yes, with evidence quotes
Incumbent PIIDs and contractors	No	Yes
Page limits, fonts, margins, submission portal	No	Yes
CMMC / NIST / FedRAMP / ITAR / IL levels	No	Yes
SBIR / STTR solicitations	No	Yes, including every open DoD topic
Only new or changed notices on scheduled runs	No	Yes, built in
AI Go/No-Go summary	No	Optional, bring your own key

---

How it works

1. Discover. Runs a full-text search on SAM.gov's public search API for each keyword, filtered by notice type, NAICS, PSC, set-aside and date. No API key is required. If you add a free SAM.gov API key, the official Opportunities v2 API is used as a second source. SBIR/STTR solicitations come from the SBIR.gov API. When SBIR.gov is offline, the Actor falls back automatically to Grants.gov SBIR/STTR funding opportunities. Every open and pre-release DoD SBIR/STTR topic (Army, Navy, Air Force, Space Force, DARPA, OSD, SOCOM and more) comes straight from DSIP, the DoD SBIR/STTR Innovation Portal, as one item per topic.
2. Skip what you've already seen (for schedules). Turn on deduplicateAcrossRuns and a persistent Key-Value Store (GOVCON-SEEN-IDS) remembers what earlier runs delivered. Each scheduled run then only returns, and only charges for, notices that are new or have changed since (for example, an amendment with a new PWS).
3. Deep-parse. Attachments are streamed into memory (never written to disk) with a 15 MB per-file safety cap. The Actor reads the first 60 pages of each PDF, all paragraph and table text of each DOCX, and the PDF/DOCX files inside ZIP packages. The SOW, PWS, RFP and Section L/M files are downloaded first.
4. Extract. A deterministic regex and heuristic engine reads the notice description plus every parsed file and returns structured signals with short evidence quotes, so you can check each one in seconds.
5. (Optional) Summarize. If you add your own Gemini, OpenAI or Anthropic key, an LLM writes a Go/No-Go call, an executive summary and a mandatory-requirements checklist for each solicitation.

---

Quick start

Zero config: open the Actor and click Start.

Typical capture-team setup (daily schedule, small-business set-asides in your NAICS codes):

{
    "keywords": ["machine learning", "computer vision", "sensor fusion", "digital engineering"],
    "naicsCodes": ["541511", "541512", "541715", "541330"],
    "noticeTypes": ["o", "k", "p", "r"],
    "setAsideTypes": ["SBA", "SDVOSBC", "8A"],
    "postedWithinDays": 3,
    "maxItems": 100,
    "includeSbir": true,
    "parseAttachments": true,
    "maxAttachmentsPerNotice": 5,
    "deduplicateAcrossRuns": true
}

Schedule it daily under Schedules, connect the Slack, email or Google Sheets integration, and your team gets only the new RFPs each morning, already qualified.

---

Input

Field	Type	Default	What it does
keywords	string[]	["artificial intelligence", "sensor fusion", "autonomous"]	One SAM.gov full-text search per phrase (all words must match). Results are merged and de-duplicated.
naicsCodes	string[]	none	Only notices with these NAICS codes.
pscCodes	string[]	none	Only notices with these Product Service Codes.
noticeTypes	string[]	["o","p","k","r"]	o Solicitation, p Presolicitation, k Combined Synopsis/Solicitation, r Sources Sought, s Special Notice, i Intent to Bundle, u J&A, a Award.
setAsideTypes	string[]	none	SBA, SBP, 8A, 8AN, HZC, HZS, SDVOSBC, SDVOSBS, WOSB, WOSBSS, EDWOSB, EDWOSBSS, VSA, VSS.
postedWithinDays	integer	7	Notices posted or modified in the last N days (1 to 365).
maxItems	integer	25	Hard cap on solicitations returned and charged per run (1 to 1000).
includeSbir	boolean	true	Also return open SBIR/STTR solicitations and DoD SBIR/STTR topics (open and pre-release) that match your keywords.
parseAttachments	boolean	true	Download and parse PDF/DOCX/ZIP attachments. Turn off for a fast metadata-only run.
maxAttachmentsPerNotice	integer	5	Files parsed per solicitation, highest priority first.
includeAttachmentText	boolean	false	Adds the cleaned full text of each parsed file (up to 100,000 characters per file). Useful for RAG and AI agents.
deduplicateAcrossRuns	boolean	false	Turn on for scheduled runs: only return and charge for new or updated notices. Each distinct set of filters keeps its own history.
llmProvider	string	none	none, gemini, openai or anthropic.
llmApiKey	secret	none	Your key for the selected provider. Stored encrypted and only sent to that provider.
llmModel	string	provider default	Optional override. Defaults are gemini-2.5-flash, gpt-4o-mini and claude-haiku-5-5.
samApiKey	secret	none	Optional free SAM.gov API key. Adds the official Opportunities v2 API as a second discovery source.
proxyConfiguration	object	no proxy	Enable Apify Proxy only if you run very large searches and start seeing HTTP 429 or 403 responses.

---

Output

Each solicitation is one dataset item. Download it as JSON, CSV, Excel or HTML, or read it through the API. The Overview view gives a one-row-per-RFP table and the Compliance & capture intel view focuses on frameworks, PIIDs and contacts.

Illustrative example (abridged; the values below are invented for this README and are not from a real notice):

{
    "source": "sam.gov",
    "notice_id": "0f3c1a9b2d4e4f6a8b1c2d3e4f5a6b7c",
    "solicitation_number": "W56KGY-26-R-0042",
    "title": "AI-Enabled Multi-Sensor Fusion Analytics Support",
    "url": "https://sam.gov/opp/0f3c1a9b2d4e4f6a8b1c2d3e4f5a6b7c/view",
    "notice_type": "Combined Synopsis/Solicitation",
    "agency": "DEPT OF DEFENSE",
    "sub_agency": "DEPT OF THE ARMY",
    "office": "W6QK ACC-APG",
    "naics_codes": ["541715"],
    "psc_code": "AC12",
    "set_aside": "Total Small Business Set-Aside (FAR 19.5)",
    "posted_date": "2026-10-14T15:02:11Z",
    "response_deadline": "2026-11-13T21:00:00Z",
    "place_of_performance": "Aberdeen Proving Ground, MD, UNITED STATES",
    "points_of_contact": [
        { "name": "Contracting Specialist", "email": "contracting.specialist.civ@army.mil", "phone": "(410) 555-0142", "role": "primary" }
    ],
    "attachments_parsed": 3,
    "attachments": [
        { "filename": "Attachment 1 - PWS.pdf", "file_type": "pdf", "status": "parsed", "page_count": 42, "pages_parsed": 42, "char_count": 118220 },
        { "filename": "Section L and M.docx", "file_type": "docx", "status": "parsed", "page_count": 11, "char_count": 30411 },
        { "filename": "DD254 Draft.pdf", "file_type": "pdf", "status": "parsed", "page_count": 4, "char_count": 9902 },
        { "filename": "Pricing Workbook.xlsx", "file_type": "other", "status": "skipped", "status_reason": "Unsupported file type (PDF, DOCX and ZIP packages are parsed)" }
    ],
    "signals": {
        "clearance_level": "TS/SCI",
        "clearance_evidence": ["[Attachment 1 - PWS.pdf] ...all contractor personnel shall possess a final TS/SCI clearance prior to start of performance..."],
        "polygraph_required": false,
        "has_dd254": true,
        "evaluation_type": "Best Value Tradeoff",
        "evaluation_evidence": ["[Section L and M.docx] ...award will be made using a best value tradeoff source selection process in accordance with FAR 15.101-1..."],
        "referenced_contract_numbers": ["W911QX21C0017"],
        "incumbent_contractors": ["Example Analytics Group, LLC"],
        "security_and_compliance_frameworks": ["CMMC Level 2", "NIST SP 800-171", "DFARS 252.204-7012", "IL5"],
        "submission_constraints": {
            "page_limits": [
                { "pages": 25, "context": "[Section L and M.docx] ...Volume I, Technical, shall not exceed twenty-five (25) pages..." }
            ],
            "formatting_requirements": ["Font size: 12 pt", "Font family: Times New Roman", "Margins: 1 inch", "Paper: 8.5 x 11 in"],
            "delivery_methods": ["PIEE Solicitation Module"]
        },
        "extracted_contacts": ["contracting.specialist.civ@army.mil", "technical.poc.civ@army.mil"],
        "extracted_phone_numbers": ["(410) 555-0142"]
    },
    "llm": {
        "provider": "gemini",
        "model": "gemini-2.5-flash",
        "status": "ok",
        "go_no_go": "REVIEW",
        "go_no_go_rationale": "Strong technical fit, but TS/SCI staffing at start of performance is a hard gate.",
        "executive_summary": "The Army is buying multi-sensor fusion analytics support at Aberdeen Proving Ground...",
        "mandatory_requirements": ["Final TS/SCI for all personnel at start", "CMMC Level 2", "Technical volume max 25 pages", "Submit via PIEE by 13 Nov 2026"],
        "key_risks": ["Incumbent referenced under W911QX21C0017"],
        "recommended_next_steps": ["Confirm cleared staffing plan", "Request incumbent contract details through FOIA"]
    },
    "deep_parsed": true,
    "charged_event": "solicitation-deep-parsed",
    "change_status": "new",
    "scraped_at": "2026-10-16T12:00:05Z"
}

Signal reference

Signal	Possible values
clearance_level	TS/SCI with Poly, TS/SCI, Top Secret, Secret, Confidential, Public Trust, Unclassified / None Stated (highest level found wins)
evaluation_type	LPTA - Lowest Price Technically Acceptable, Best Value Tradeoff, HTRO - Highest Technically Rated Offeror, Unspecified
referenced_contract_numbers	Uniform PIIDs such as W911NF21C0001, FA865019C1234, N0001418D1234, 47QTCA19D00AB, plus GSA schedules such as GS-35F-0119Y. The notice's own solicitation number is excluded.
security_and_compliance_frameworks	CMMC and FedRAMP (with the level when the documents name a single one), NIST SP 800-171, NIST SP 800-53, ITAR, EAR, DFARS 252.204-7012, DFARS 252.204-7019/7020 (SPRS), DFARS 252.204-7021 (CMMC), IL2, IL4, IL5, IL6, FISMA, FIPS 140, CUI, Section 508, Zero Trust, CJIS, HIPAA, SOC 2
submission_constraints.delivery_methods	PIEE Solicitation Module, SAM.gov, Email, Vulcan, DSIP (DoD SBIR/STTR Innovation Portal), Grants.gov Workspace, Research.gov, NSPIRES (NASA), eRA Commons / ASSIST (NIH), GSA eBuy, FedConnect, DoD SAFE, Physical delivery (hand carry / courier)

Every run also writes a run summary to the OUTPUT record of the default Key-Value Store. It holds counts of discovered, skipped, scraped and deep-parsed notices, per-source status and warnings. Agents and pipelines can use it to check a run at a glance.

---

Pricing (Pay-Per-Event)

You only pay for solicitations that are delivered to your dataset:

Event	When it is charged
solicitation-scraped	A solicitation is delivered with notice metadata and signals from the description, but no attachment text was available (no attachments, controlled files, or parseAttachments is off).
solicitation-deep-parsed	At least one PDF or DOCX attachment was downloaded and its text was parsed.

See the Pricing tab for the current price of each event.

• With deduplicateAcrossRuns on, notices delivered in earlier runs are skipped for free, so a daily schedule only bills for what is new.
• Your run's maximum charge is respected. The Actor checks the remaining budget before each item and stops cleanly when the budget runs out.
• LLM costs, if you enable the AI pass, go to your own provider account.

---

Use it from AI agents (MCP)

Every input field has a default and a plain-language description, so AI agents can call this Actor without any setup.

Hosted Apify MCP server (Claude, ChatGPT, Cursor, VS Code and any client that supports remote MCP). Add this server URL:

https://mcp.apify.com?tools=AIMindTeams/govcon-rfp-deep-parser

Local stdio server (for example, Claude Desktop claude_desktop_config.json):

{
    "mcpServers": {
        "govcon-rfp": {
            "command": "npx",
            "args": ["-y", "@apify/actors-mcp-server", "--tools", "AIMindTeams/govcon-rfp-deep-parser"],
            "env": { "APIFY_TOKEN": "YOUR_APIFY_TOKEN" }
        }
    }
}

Example prompts:

• "Find Secret-or-lower AI and computer vision RFPs posted this week with small-business set-asides, and list page limits and due dates."
• "Which new sensor fusion solicitations reference an incumbent contract? Give me the PIIDs and contracting officer emails."
• "Pull this week's SBIR/STTR topics on autonomy and summarize the mandatory requirements."

Tip for agents: set includeAttachmentText to true when the agent needs to answer detailed questions about the documents themselves.

---

Use it from code

Python (pip install apify-client):

from apify_client import ApifyClient

client = ApifyClient("YOUR_APIFY_TOKEN")
run = client.actor("AIMindTeams/govcon-rfp-deep-parser").call(
    run_input={
        "keywords": ["counter-UAS", "sensor fusion"],
        "setAsideTypes": ["SBA"],
        "postedWithinDays": 14,
        "maxItems": 50,
    }
)
for item in client.dataset(run["defaultDatasetId"]).iterate_items():
    signals = item["signals"]
    print(item["solicitation_number"], signals["clearance_level"], signals["evaluation_type"], item["url"])

cURL (run and get the items in one call):

curl -X POST "https://api.apify.com/v2/acts/AIMindTeams~govcon-rfp-deep-parser/run-sync-get-dataset-items?token=YOUR_APIFY_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"keywords": ["artificial intelligence"], "maxItems": 10}'

---

FAQ

Do I need a SAM.gov account or API key? No. Discovery uses SAM.gov's public search API, and attachments marked public download without a login. A free SAM.gov API key is optional and only adds a second discovery source.

Which attachments are parsed? Public PDF, DOCX and ZIP files (PDF/DOCX inside ZIPs). Each file is capped at 15 MB, and only the first 60 pages of a PDF are read, which covers the SOW and evaluation sections while skipping long FAR clause appendices. Controlled attachments that need a SAM.gov login and explicit access approval are listed with status skipped. Spreadsheets, images and legacy .doc files are listed as skipped too.

What about scanned PDFs? The Actor doesn't run OCR. Image-only PDFs are listed with status failed and the reason "no extractable text layer". If a notice has no readable attachments, it is charged as solicitation-scraped, not as deep-parsed.

How accurate are the signals? The engine is deterministic: the same text always gives the same result. It is tuned for precision and checks for negations ("no security clearance is required", "tradeoffs are not permitted"). Every clearance and evaluation call includes evidence quotes with the file name, so you can verify it quickly. Treat the output as a fast first read, not a replacement for reading Section L and M before you bid.

How does deduplication work with several saved tasks? History is stored in the GOVCON-SEEN-IDS Key-Value Store under a key derived from your filters (keywords, NAICS, PSC, notice types, set-asides, SBIR on/off). Two tasks with different filters never hide results from each other. It is off by default, so one-off searches and AI agents get every match on every run; turn it on in your scheduled task.

SBIR.gov says its API is under maintenance. Do I still get SBIR results? Yes. The Actor tries SBIR.gov first and falls back automatically to Grants.gov SBIR/STTR funding opportunities (NIH, NSF, USDA, DOE and other agencies), including their NOFO PDFs. DoD topics always come from DSIP, so they are covered even when SBIR.gov is down. Each DoD topic is one item with its objective, description, phases, keywords, CMMC level, ITAR flag, Q&A dates and topic contact. The run summary shows which source was used. Grants.gov sometimes blocks automated downloads of NOFO PDFs. When that happens the opportunity is still delivered with its full synopsis and eligibility text, it is charged as metadata-only, and the attachment entry explains why.

Is this legal? The Actor only reads publicly posted federal procurement data and public attachments. It does not log in or bypass access controls. You are responsible for using the data in line with SAM.gov's terms of use.

---

Changelog

• 1.1: DoD SBIR/STTR topics from DSIP (source dsip): every open and pre-release topic as its own item, with objective, phases, cost limits, CMMC level, ITAR flag and topic contacts.
• 1.0: First release. SAM.gov website search and optional v2 API discovery, SBIR.gov with Grants.gov fallback, in-memory PDF/DOCX/ZIP parsing, deterministic GovCon signal engine, cross-run deduplication, Pay-Per-Event billing and optional BYOK LLM Go/No-Go enrichment.