Agent Economy · Model Context Protocol · Bazantic Gateway

Agent-to-Agent Question Bank API & MCP

CertScroll is built for both human mastery and autonomous AI agents. Expose 2,050+ syllabus-verified exam questions, cognitive diagnostic matrices, and blueprint mock drills directly into Claude Desktop, Cursor, LangChain, or autonomous evaluation workflows via the Bazantic MCP Gateway.

⚡ Open on Bazantic Marketplace →📄 OpenAPI 3.1 Spec 🤖 MCP Tools Manifest

Architecture & the dual-surface system

🤖

Model Context Protocol (MCP)

Standardized tool execution for LLMs. Exposes 5 primitives: get_questions, grade_question, generate_mock_exam, get_concept_blueprint, and get_matrix_coverage.

⚡

Bazantic Gateway & x402-mpp

Reverse-proxied via Bazantic.com on Fly.io. Handles rate limits, authentication, usage metering, and per-query micro-USDC payments on Base L2 via standard HTTP 402 headers.

🎯

Blind LLM Evaluation

Prevent evaluation data leakage with maskAnswers: true. Candidate models formulate chain-of-thought independently before automated grading against official syllabus leaf citations.

[Human Learner] ──────────► study.wanessalabs.com 
                              └─ Astro 5 SSG + Cloudflare Pages (100% Client-Side Dexie / IndexedDB)

[AI Agent / IDE] ─────────► ai-cert-questions.bazgateway.com 
                              └─ Bazantic Edge Gateway (Rate limiting, auth, x402-mpp settlement)
                                   ├─ JSON-RPC 2.0 /mcp (Model Context Protocol endpoint)
                                   └─ REST API /api/v1/* (Questions, Grade, Mock Exams)

Client installation & agent setup

Add the CertScroll MCP server to your AI environment. Agents connect directly over JSON-RPC 2.0 or SSE.

{
  "mcpServers": {
    "certscroll": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-fetch",
        "https://ai-cert-questions.bazgateway.com/mcp"
      ]
    }
  }
}

Save to %APPDATA%\Claude\claude_desktop_config.json (Windows) or ~/Library/Application Support/Claude/claude_desktop_config.json (macOS).

{
  "mcpServers": {
    "certscroll": {
      "url": "https://ai-cert-questions.bazgateway.com/mcp"
    }
  }
}

Save to .cursor/mcp.json in your project root or add via Cursor Settings → Features → MCP Servers.

How agents pay: x402 Machine Payment Protocol (MPP)

Autonomous agents don't have credit cards, company email addresses, or browser sessions. CertScroll supports the standard HTTP 402 Payment Required protocol with on-chain micro-settlement on Base.

1. Agent Request (Unauthenticated):
   GET /api/v1/questions?cert=aws-mla-c01 HTTP/1.1
   Host: ai-cert-questions.bazgateway.com

2. Gateway Challenge:
   HTTP/1.1 402 Payment Required
   x402-price: 0.005
   x402-asset: USDC
   x402-network: base
   x402-recipient: 0x2A5b...4F12

3. On-Chain Settlement:
   Agent signs & broadcasts $0.005 micro-USDC transfer on Base (<$0.001 gas, sub-2s finality).

4. Agent Retry with Cryptographic Proof:
   GET /api/v1/questions?cert=aws-mla-c01 HTTP/1.1
   Host: ai-cert-questions.bazgateway.com
   Authorization: x402 0x9f8e... (txHash or signed EIP-712 payment receipt)

5. Gateway Validation & Payload Delivery:
   HTTP/1.1 200 OK
   Content-Type: application/json
   [ Verified Syllabus Questions Payload ]

Developer Alternative: Pre-Funded Bearer API Key

If your agent runtime does not yet have an active on-chain wallet (e.g. Claude Desktop or Cursor), create an API key on the Bazantic Marketplace. Include Authorization: Bearer <key> in your request. Bazantic provides 10 free queries/day for schema exploration.

Five Core MCP Primitives

Tool Name	Parameters	Role & Description	Unit Pricing
`get_questions`	`cert`, `concept`, `difficulty`, `limit`, `maskAnswers`	Retrieve syllabus-cited questions across 10 certs. Supports blind evaluation masking via `maskAnswers: true`.	$0.005 / call
`grade_question`	`questionId`, `selectedOption`, `agentReasoning`	Grade answer choice (0–3) and evaluate chain-of-thought against leaf citations and distractor rationales.	$0.002 / call
`generate_mock_exam`	`certId`, `questionCount` (15, 25, 50), `maskAnswers`	Generate domain-stratified mock exam batches dynamically weighted by official certification blueprint proportions.	$0.05 / exam
`get_concept_blueprint`	`slug`	Retrieve knowledge graph node, overlaps across 10 certifications, and related concept prerequisites.	$0.01 / call
`get_matrix_coverage`	(none)	Query real-time matrix coverage stats across all 765 reachable cells and 4 cognitive difficulty tiers.	$0.001 / call

Ready-to-use agent test recipes

Paste these prompts into Claude Desktop, Cursor, or your autonomous eval runner to test the live endpoint:

Recipe 1 · Discovery

Tool Discovery & Schema Verification

Prompt to confirm your agent correctly discovers all 5 MCP tool definitions and schemas:

Connect to the CertScroll MCP server at https://ai-cert-questions.bazgateway.com/mcp.
List all available tools and inspect their input schemas. Verify that you can see get_questions, grade_question, generate_mock_exam, get_concept_blueprint, and get_matrix_coverage.

Recipe 2 · Blind Eval

Blind Scenario Solving & Grading

Prompt to test zero-contamination evaluation and verify chain-of-thought against syllabus citations:

1. Call get_questions with cert: "aws-mla-c01", difficulty: "application", limit: 3, maskAnswers: true.
2. Read each scenario, analyze trade-offs between the 4 options, output your chain-of-thought reasoning, and state your final answer (0, 1, 2, or 3).
3. Call grade_question with questionId and selectedOption to verify against the official syllabus leaf citation.

Recipe 3 · Mock Exam

Stratified Mock Exam Benchmark

Prompt to run a full 15-question domain-weighted benchmark:

Call generate_mock_exam with certId: "anthropic-claude-certified-architect", questionCount: 15, maskAnswers: true.
Simulate answering the mock exam, submit your choices to grade_question, and compute your final percentage score against the 70% passing benchmark.

Recipe 4 · Knowledge Graph

Cross-Syllabus Overlap Traversal

Prompt to inspect how core AI concepts connect across providers:

Call get_concept_blueprint with slug: "lora".
Summarize how LoRA (Low-Rank Adaptation) is tested across AWS, Databricks, and GCP, and explain its prerequisite concepts in the knowledge graph.

In-browser API & MCP explorer

Connect Claude Desktop, Cursor, Goose, or autonomous agent evaluation harnesses to CertScroll through the Bazantic Gateway:

{
  "mcpServers": {
    "certscroll": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-fetch",
        "https://ai-cert-questions.bazgateway.com/mcp"
      ]
    }
  }
}

Connect to the CertScroll MCP server at https://ai-cert-questions.bazgateway.com/mcp.
1. Call get_questions with cert: "aws-mla-c01", difficulty: "application", limit: 3, maskAnswers: true.
2. Read each scenario, analyze trade-offs, and state your final answer (0, 1, 2, or 3).
3. Call grade_question with the questionId and selectedOption to check against official syllabus leaf citations and explanations.

Five Core Model Context Protocol (MCP) Primitives:

get_questions

Filter by certification, cognitive tier, or concept. Supports maskAnswers: true for zero-leakage blind evals.

grade_question

Grade choice (0–3) and evaluate chain-of-thought against official syllabus leaf citations and distractor rationales.

generate_mock_exam

Generate blueprint-stratified mock exam batches (15, 25, 50 questions) weighted by official exam domains.

get_concept_blueprint

Retrieve knowledge graph node, overlaps across 10 certifications, and related concept prerequisites.

get_matrix_coverage

Query real-time matrix coverage stats across all 765 reachable cells and 4 cognitive difficulty tiers.

Developer & agent pricing

✓ Full interactive human web app & PWA
✓ 10 free API queries / day
✓ OpenAPI 3.1 schema download
✓ Spaced repetition tracking

✓ Full Model Context Protocol (MCP) server access
✓ Blueprint-stratified mock exam generator ($0.05 / exam)
✓ Syllabus domain citations & distractor analyses
✓ Metered via Bazantic API Key or x402 headers on Base

Access on Bazantic Marketplace →

✓ Full 2,050+ question raw dataset export (JSONL)
✓ Automated LLM grading benchmark pipeline
✓ Custom certification domain expansion
✓ Private on-premises MCP container

API Endpoint Catalog (v1)

Method	Endpoint	Description	Format
GET	`/api/v1/questions`	List question bank with filters for cert, domain, tier, and limit.	JSON
GET	`/api/v1/questions/:id`	Retrieve a single question card by ID with explanation & metadata.	JSON
GET	`/api/v1/certs`	Catalog of 10 certification tracks with domain weights & URLs.	JSON
GET	`/api/v1/concepts`	Overlapping AI/ML concepts and prerequisites.	JSON
GET	`/api/v1/matrix`	Shared concept overlap matrix across all certifications.	JSON
GET	`/api/v1/openapi.json`	OpenAPI 3.1.0 specification for code generators and MCP gateways.	JSON
GET	`/api/v1/mcp-manifest.json`	Model Context Protocol server declaration and Bazantic gateway config.	JSON