cert-atlas-mcp

An MCP server over Cert Atlas — the open index of certification exam blueprints. It lets any AI assistant (Claude, Cursor, …) search exams, pull a full structured blueprint, and compare certifications, with a free practice link on every result.

Cert Atlas publishes machine-readable blueprints for 1,580 exams across 222 certifying bodies — exam domains and topic weights, passing scores, question counts, durations, prerequisites, retake/renewal rules, languages, and official source links. This server exposes that dataset to agents over MCP. Read-only, public data, no API key.

Tools

| Tool | What it does | |------|--------------| | search_exams | Keyword search across all 1,580 blueprints (vendor, exam code, cert name, topic). Optional certifying_body / vendor_slug filter. Returns matches with id, domain count, question count, duration, and a practice link. | | get_exam_blueprint | Full blueprint for one exam (by id, code, or name): domains + weights, passing score, format, price, languages, prerequisites, retake/renewal policy, and official source URL. | | compare_exams | Compare 2–8 exams side by side: question count, duration, passing score, price, validity, domain count. | | list_certifying_bodies | All 222 certifying bodies with exam counts (optional contains filter) — useful to scope a search. |

Resources & prompts

• Resources: cert-atlas://index (the master index) and cert-atlas://exam/{exam_id} (a single blueprint) — agents can browse or attach the data directly.
• Prompts: study-plan (arg: cert) builds a topic-weighted plan grounded in the real blueprint; compare-certs (arg: certs) drives a structured comparison.

Practice-link funnel

Every exam returned by any tool carries its free QuizForge practice link with UTM attribution — ?utm_source=mcp&utm_medium=cert_atlas&utm_campaign=<tool> (existing query preserved) — so the server doubles as a measurable referral funnel. Links are factual and singular (one per exam); the value to the user is a real free practice exam.

Install

npm install
npm run build

Use with Claude Desktop / Claude Code

Add to your MCP config (claude_desktop_config.json, or via claude mcp add):

{
  "mcpServers": {
    "cert-atlas": {
      "command": "node",
      "args": ["/absolute/path/to/cert-atlas/mcp/dist/index.js"]
    }
  }
}

Or, once published to npm (works in Claude Desktop/Code and Cursor — .cursor/mcp.json):

{
  "mcpServers": {
    "cert-atlas": { "command": "npx", "args": ["-y", "cert-atlas-mcp"] }
  }
}

Example prompts

• "What domains are on the AWS Solutions Architect Associate exam, and how is it weighted?"
• "Compare CompTIA Security+ and CySA+ — which is harder and which is cheaper?"
• "What are the prerequisites and passing score for CISSP?"
• "What nursing certifications does Cert Atlas cover?"

How it works

The server reads the Cert Atlas dataset, preferring local files over the network:

1. A bundled snapshot of index.json (+ vendors.json) ships in the package → instant, offline search with a tiny install.
2. Individual blueprints are lazy-fetched from raw.githubusercontent.com/hans6883/cert-atlas/master on demand and cached, so the install stays small and data stays fresh.
3. When run inside the cert-atlas repo, it reads the canonical ../data/ directly (always in sync).

The bundled snapshot is regenerated from ../data on every npm run build (see scripts/copy-data.mjs).

Environment variables (all optional)

| Var | Default | Purpose | |-----|---------|---------| | CERT_ATLAS_LOCAL | — | Path to a cert-atlas repo checkout; reads its data/ dir (used by the VPS/remote deployment). | | CERT_ATLAS_DATA_DIR | repo/bundled data/ | Force a specific data directory. | | CERT_ATLAS_RAW_BASE | https://raw.githubusercontent.com/hans6883/cert-atlas/master | Override the remote dataset base (e.g. a fork or mirror). |

Remote HTTP build

The same four tools are also served over the MCP Streamable HTTP transport, for hosting with no client install:

npm run build
npm run start:http        # PORT (default 3000), HOST (default 0.0.0.0)
# POST /mcp   — MCP endpoint
# GET  /health — liveness probe -> {"ok":true}

It's stateless (a fresh server per request), so it scales horizontally with no session store. A Dockerfile is included:

docker build -t cert-atlas-mcp .
docker run -p 3000:3000 cert-atlas-mcp

Put it behind your reverse proxy (TLS) and point HTTP-capable MCP clients at https://<host>/mcp.

Live endpoint: https://mcp.quizforge.ai/mcp (health: https://mcp.quizforge.ai/health) — add it as a custom/remote connector in any HTTP-capable MCP client, no install.

Publishing

npm (stdio package):

npm login
npm publish --access public      # runs the build via prepublishOnly

Official MCP registry (registry.modelcontextprotocol.io) — uses server.json; package.json carries the matching mcpName for ownership verification:

# one-time: install the publisher CLI, then authenticate via GitHub
mcp-publisher login github
mcp-publisher publish            # reads ./server.json

smithery.yaml lets Smithery list it. Other directories (Glama, PulseMCP, mcp.so) auto-index from npm + GitHub once published.

License

MIT — see the repo LICENSE. Cert Atlas indexes publicly available exam structure only (no proprietary questions); blueprint content belongs to the respective certifying bodies.