Menu
Skills Pricing What's MCP? Docs
Sign in

Connector Documentation

Proofwork provides certified AI skills as Model Context Protocol (MCP) servers. Connect once in your AI assistant of choice — Claude, ChatGPT, or any MCP client — and use the tools in natural conversation.

Overview

Every Proofwork skill is a narrow, benchmarked tool designed for high-stakes work where being approximately right isn't good enough. Each skill ships with published accuracy scores, a known failure taxonomy, and a frozen reference dataset — so you can audit its behavior instead of trusting a black box.

Available today

HTS tariff classification — 90.5% validated accuracy on a 100-case benchmark. Six tools for structured heading → subheading → leaf-code workflows, backed by 20,000+ CBP ruling precedents.

Coming soon

AD/CVD exposure screening, duty stack + landed cost, USMCA/FTA pre-check, denied-party screening. All shipping as separate certified MCP servers with their own benchmarks.

How to connect

Proofwork's MCP server lives at one public URL. Every supported client points at the same endpoint:

https://mcp.proofworklabs.com/mcp

Claude (web, desktop, or Claude Code)

In Claude, open Settings → Connectors → Add custom connector, paste the URL above, and complete sign-in. Claude will discover the tools automatically.

ChatGPT

In ChatGPT, enable developer mode under Settings → Apps & Connectors → Advanced settings, then add the connector from Settings → Connectors → Create. ChatGPT Apps are available on all plans as of November 2025.

Other MCP clients

Any client that supports remote MCP servers with OAuth or bearer-token auth will work. For programmatic integrations (Python MCP client library, curl, the Anthropic SDK with tool_use), see the Developer tab on the HTS get-started page.

Authentication & free tier

Proofwork uses standard OAuth 2.0 (Authorization Code + PKCE) for interactive clients and API keys for programmatic access. When you first connect, your AI assistant will open a sign-in page hosted at proofworklabs.com where you can sign in with Google, Microsoft, or email/password. After that, the connection is persistent — you won't be prompted again on subsequent tool calls.

Free tier: 1,000 calls

Every new account gets 1,000 MCP tool uses for free across all Proofwork skills. No credit card required. When you run out, you can subscribe at proofworklabs.com/pricing for unlimited access.

If your company email domain matches an organization already on Proofwork, you'll auto-join it and share the company's subscription. No invite flow needed for corporate accounts.

Worked examples

Below are three real prompts showing how the HTS classification skill handles different product types. Expected tool call sequences shown for transparency — in practice, your AI assistant orchestrates these calls automatically.

1

Simple consumer product

Tests clean heading + specific leaf code selection

Prompt:

"What's the HTS code for a 100% cotton men's t-shirt manufactured in Vietnam?"

Expected tool sequence:

  1. identify_heading_payload → picks heading 6109 (T-shirts, singlets)
  2. select_subheading_payload → picks 6109.10 (of cotton)
  3. select_leaf_payload → picks 6109.10.0012 (men's or boys', knit)

Result: 6109.10.0012 — grounded in CBP rulings for cotton knit apparel, with chapter notes and precedent evidence shown at each step.

2

Industrial product with numeric attributes

Tests breakpoint extraction (dimensions, material grade)

Prompt:

"Classify M10 stainless steel hex bolts, 50mm length, for industrial fastening."

Expected tool sequence:

  1. identify_heading_payload → heading 7318 (screws, bolts, nuts)
  2. select_subheading_payload → 7318.15 (other screws and bolts)
  3. select_leaf_payload → applies diameter breakpoint (M10 = 10mm), picks the matching hex bolt leaf
  4. verify_code_payload → compares against sibling codes (cap screws, studs) to confirm hex-bolt match

The skill correctly extracts the M10 → 10mm diameter mapping and applies the 6mm/8mm/12mm breakpoints to reach the right leaf code. Material-specific (stainless) siblings are evaluated explicitly.

3

Ambiguous product requiring material-vs-function tiebreak

Tests GRI 3(b) essential-character reasoning

Prompt:

"I'm importing a silicone kitchen spatula with a bamboo handle. What's the correct HTS code?"

Expected tool sequence:

  1. identify_heading_payload → returns multiple candidates: 3924 (plastic kitchenware), 4419 (wood tableware), 7323 (iron kitchenware, not applicable)
  2. The skill evaluates essential character per GRI 3(b): silicone provides the functional surface, bamboo is the handle. Silicone wins.
  3. select_subheading_payload → picks 3924.10 (kitchenware of plastics)
  4. select_leaf_payload → picks the kitchen utensil leaf with reasoning shown

Critical behavior: the skill shows its reasoning. If two candidate headings both have support, it surfaces the tiebreaker (here, GRI 3(b)) so you can audit the decision — rather than silently picking one.

Want to see the full benchmark? Every example above comes from real product types in our 100-case evaluation set. See the HTS skill deep-dive for the complete methodology, failure taxonomy, and CROSS ruling coverage.

Limits & availability

Tool response size

MCP tool results are capped at 25,000 tokens per call. Proofwork's tools return compact structured payloads well below this limit.

Tool timeouts

MCP clients generally enforce a 5-minute timeout per tool call. Proofwork tools typically respond in under 2 seconds.

Hosting

Proofwork's MCP server runs on Google Cloud Run in us-central1 with managed TLS. No customer data is persisted — tool calls are stateless.

Updates

Reference datasets (CBP rulings, HTSUS schedule) are refreshed on a published cadence. Version numbers and refresh dates are exposed in each skill's deep-dive page.

Support

Questions, bug reports, or integration help:

Ready to connect?

Add Proofwork to your AI assistant in under a minute.

Get started with HTS