Measured, reliable mineral evidence
Know when mineral
evidence is trustworthy.
Turn a lab certificate, drill log, or assay report into structured, provenance-linked JSON with measurable evidence reliability — so an AI agent can tell what it can trust, what it cannot, and when to refuse. Two parties can settle a high-stakes claim without having to trust each other, or us.
The U.S. holds decades of unsearchable critical-mineral data — assay certificates, drill logs, and state-survey archives locked in PDFs and legacy formats. AssayChain builds the verifiable data-integrity layer that turns that backlog into actionable intelligence without new drilling — and we are building the evidence-reliability layer: fidelity, completeness, consistency, provenance, and evidentiary sufficiency — so an agent knows not just what a document says, but whether the evidence is sufficient to trust and act on.
Source integrity · extraction fidelity · schema validity · cross-extraction agreement · unit consistency · provenance · evidence conflicts.
That a laboratory result is scientifically correct · that a deposit is economic · that a filing is compliant · that a geological interpretation is correct. Provenance verifies the record; evidence reliability evaluates the claim — Qualified Person review remains required.
Who it is for
Prove assay data to buyers, investors, and regulators.
Verify supplier content claims.
Verify reserve and ESG disclosures.
Digitize legacy data at volume.
Best of the Best • Mineral Data (Last 10 Days)
Live proof the pipeline works: each row is a real assay or geology document we structured, graded, and attested in the last 10 days — the demo that shows how the nation's legacy critical-mineral data becomes verifiable intelligence without new drilling. Grouped by location. Raw JSON for LLMs →
| Document | Location | Grade | Samples | Date | Highlights | |
|---|---|---|---|---|---|---|
| Loading recent curated data… | ||||||
What we provide
Structure it. Query it. Pay per file.
AssayChain's core product: turn unstructured assay reports into structured, graded JSON, and query 20 USGS critical minerals through a clean REST + MCP API. Credit card for humans; machine-to-machine micropayment for AI agents. Additional intelligence endpoints (risk scores, compliance checklists) are offered as experimental tools — helpful for research, never a substitute for professional review.
Assay Extract API
Upload any public PDF, image, CSV, or text assay report. AI-powered extraction returns every sample row as structured JSON with an A–D quality grade. 80% refund on Grade D. $0.10–$5.00 per job. Proven on USGS fire-assay data — 57 samples, grade A.
NI 43-101 → CSV converter tool ↗ — free utility that formats your assay tables for downstream use.
USGS Commodity Benchmarks
Official USGS mineral commodity summaries for 20 critical minerals — production, reserves, top producers, China share. Each response carries verifiable provenance with content-addressed source data. Free and rate-limited.
Experimental research tools Assistive
Enrichment, supply-chain risk, compliance checklists, ore-potential, and district reports — offered as assistive research context, never a substitute for Qualified Person review. Not part of the core reliability research.
How payment works
Machine-to-machine payment in four steps
Call any endpoint
Your agent or app sends a standard HTTP GET — no auth header needed.
Receive 402
The server returns 402 Payment Required with a machine-readable USDC offer.
Sign & send
Your x402 client signs the micropayment authorization and sends the quoted USDC amount.
Receive response
The server verifies via settlement service, settles USDC, and returns the full payload with attestation UID and content-addressed CID.
Why not use USGS directly?
USGS PDFs vs. AssayChain API
The USGS Mineral Commodity Summaries are the industry standard. But they're published as PDFs — not an API. Here's what AssayChain adds.
| USGS PDFs | AssayChain | |
|---|---|---|
| Format | PDFs you must download, parse, and extract manually | Structured JSON via REST — HTTP GET, returns ready-to-use data |
| Update cadence | Annual reports (MCS released each January) | Same USGS data + daily price snapshots for 8 commodities (FRED/LME) |
| Data enrichment | Raw numbers only | AI-enriched: grade benchmarks, typical ranges, processing notes, deposit types |
| Supply-chain risk | Not available — raw production tables only | Risk scores (0–100), China dependency, allied supply ratio, resilience scenario tests, ESG profiles |
| Compliance | Not available | NI 43-101 checklists, JORC Table 1 reports, DFARS/IRA/EU Battery crosschecks |
| Payment model | Free (but time cost to extract) | $0.10–$5.00 extraction · $0.10–$0.50 data · via an open micropayment protocol (USDC) or credit card. No monthly commitment. |
| Provenance | None — you trust the PDF | Every response includes a cryptographic attestation UID + content-addressed CID — verify independently |
| AI agent ready | No — PDFs require OCR and custom parsing | Yes — MCP endpoint, automatic micropayment, structured JSON schemas |
| Coverage | USGS — US-focused | Multi-jurisdiction: US DoD, EU, IEA, Japan METI criticality lists. Trade controls for China export restrictions. |
Why it matters
Mineral data is scattered and hard to use
USGS mineral commodity data is technically public, but it’s locked in PDFs and annual reports that require manual extraction. No structured API. No machine-readable format. No provenance chain from source document to delivered response.
Assay results and geological records are equally fragmented — scattered across lab certificates, drill logs, and internal reports. The next operator starts from scratch.
AssayChain structures all of it: source data is sanitized against locked schemas, stored in content-addressed storage, recorded with a verifiable attestation, and served through a single metered API. AI agents and developers get clean, verifiable mineral data — without accounts, subscriptions, or trust assumptions.
Five product surfaces: (1) USGS Mineral Commodity Summaries for 20 critical minerals, (2) curated historical US/MX mining district reports, (3) daily mineral market context bundles (spot price, BDI, WTI, FX, LME stock), (4) AI-powered structured assay extraction from PDFs/images/CSVs, and (5) company critical-minerals exposure profiles assembled by a 3-agent swarm over the USGS corpus. All data schema-locked, content-addressed, and verifiably attested.
Each source record is stored in content-addressed storage — a network where a file's address is derived from its content (so any tampering changes the address). A permanent, signed pointer to that record is then recorded in a public attestation registry. Anyone can follow that pointer and verify the original data, without trusting AssayChain.
Quick start
Test the API in 30 seconds
No wallet or account required to test. The health check and extract estimate are always free.
1. Free health check
curl https://assaychain.com/api/health
2. Trigger the x402 gate (no payment taken)
curl -i https://assaychain.com/api/benchmark/copper
HTTP 402 Payment Required is the correct response — it means the gate is working. Use an x402 client (Claude Desktop, Cursor, or x402 SDK) to pay and receive data.
3. Free extraction estimate (no payment)
curl -X POST https://assaychain.com/api/extract/estimate \
-H "content-type: application/json" \
-d '{"source_url":"https://pubs.usgs.gov/circ/1968/0592/report.pdf"}'
Returns an estimate_id, page count, quality floor grade, and USDC price quote. Use the estimate_id with POST /extract/run to pay and extract. Full API docs →
FAQ
Common questions
What is x402 and how do I pay?
x402 is an open HTTP payment protocol. When you call a paid endpoint without a payment header, the server returns HTTP 402 Payment Required with a machine-readable offer: the amount, the settlement network, and the recipient address. An x402-aware client reads that offer, signs an EIP-3009 USDC authorization with your wallet, retries the request with the X-PAYMENT header, and receives data in return.
You do not need to manually construct the payment header. Use Claude Desktop, Cursor, or the x402 TypeScript/Python SDK.
What wallet and tokens do I need?
You need a wallet with USDC. USDC is the native USDC contract at 0x833589fCD6eDb6E08f4c7C32D4f71b54bdA02913. You also need a tiny amount of ETH on Base for gas (~$0.001 per call).
To get USDC: bridge from Ethereum mainnet via bridge.base.org, or buy directly on Base via Coinbase.
What AI clients work with this API?
MCP clients (add POST /mcp URL in settings): Claude Desktop, VS Code / GitHub Copilot, Cursor, Windsurf, and any tool built on the MCP SDK. Each client has a config JSON with a "url" key pointing at https://assaychain.com/api/mcp. See the API docs for client-specific config snippets.
OpenAI Agents SDK (Python): use MCPServerStreamableHTTP pointing at the same URL.
ChatGPT Custom GPTs (Actions): requires an OpenAPI spec — contact us via the Contact page.
Custom agents / scripts: use x402-fetch (NPM) or the x402 Python package to call any gated endpoint directly with automatic payment.
What does the Extract API do and what document types does it support?
The Extract API reads assay reports — fire assay PDFs, lab certificate images, CSV data tables, or plain text — and returns every sample row as structured JSON with element, value, and unit, plus structured provenance (site, county, state, assayer, sample dates). The output schema (assay-extract-v0.2) matches the same field vocabulary as the benchmark endpoints so your AI agent can consume both with no remapping.
Supported in v0.2: PDF (digital and scanned), image (PNG/JPG/WEBP), CSV, TXT. XLSX and DOCX return a 415 error with a reason — no charge is taken for unsupported types.
See the API docs for the full step-by-step extraction workflow.
What does the extraction grade (A–D) mean?
Each extraction is scored on four signals: structural validity (did the model produce valid sample rows?), coverage (did it find at least 70% of the expected rows?), self-consistency (does a second independent extraction pass return the same numbers?), and cross-check (are the element symbols and units recognized, and within physical bounds?). Grade A = all four pass. Grade B = three of four. Grade C = two. Grade D = one or fewer.
Grade D triggers an 80% manual refund. Grade C is delivered with a warning to review manually.
How is sensitive location data protected?
Site coordinates in extraction results are rounded to ±0.1° before storage and delivery — enough precision for regional analysis but not enough to expose a specific claim. Email addresses and phone numbers in document text are stripped and replaced with [REDACTED] tokens before any data reaches the API response.
How do I verify the provenance of a response?
Benchmark and district responses include an attestation_uid and ipfs_cid. Paste the UID into base.easscan.org to inspect the full verifiable record. Fetch the IPFS CID via any public gateway to retrieve the original source data. No trust in AssayChain required. See the Provenance page for step-by-step instructions.