ThreatCluster: free CTI API and MCP server

A free threat intelligence API (100 credits a day, no card) and an MCP server over it. Ask Claude, Cursor or any MCP client what is new, who was hit, whether a CVE is exploited, and get answers from a live corpus of deduplicated incident clusters, actor and malware profiles, CVEs with KEV and EPSS, and ransomware leak-site victims.

An MCP server that gives Claude, Cursor, VS Code, Windsurf, Zed and any other MCP client live threat intelligence from ThreatCluster (setup guide): incident clusters (one deduplicated story per incident, with a threat score, timeline and extracted entities), entity profiles (actors, malware, tools, vendors, CVEs), CVE records with KEV / EPSS / exploit status, and ransomware leak-site victims.

It is a thin, auditable wrapper over the public REST API. Every tool call is one or two GETs to https://threatcluster.io/api/public/v1 with your API key; nothing else leaves your machine, and there is no telemetry.

Why this and not a feed aggregator

Most free CTI MCP servers wrap other people's indicator feeds (OTX, abuse.ch, AbuseIPDB). ThreatCluster is the corpus itself: every story is one cluster built from every source that covered it, with a threat score, a timeline and the entities extracted from the text, plus the leak-site victim list and CVE exploit status in the same API. The free key covers all of it with a 7-day lookback; paid plans extend the window. Details and the hosted endpoint (https://threatcluster.io/mcp) are at threatcluster.io/about/free-api.

Published twice from one tool spec, so both are identical:

1. Get a key

Free keys carry the five read scopes, 100 credits a day, 30 requests a minute and a 7-day lookback: https://threatcluster.io/api. Put it in THREATCLUSTER_API_KEY. If you use the tc CLI and have run tc auth login, the Python server picks that credential up automatically (keyring or the ~/.config/tc-cli/credentials file); the Node server reads the file only.

2. Add the server

Claude Code

claude mcp add threatcluster -e THREATCLUSTER_API_KEY=tc_live_... -- npx -y threatcluster-mcp
# or the Python build:
claude mcp add threatcluster -e THREATCLUSTER_API_KEY=tc_live_... -- uvx threatcluster-mcp

Claude Desktop (claude_desktop_config.json, Settings > Developer > Edit Config)

{
  "mcpServers": {
    "threatcluster": {
      "command": "npx",
      "args": ["-y", "threatcluster-mcp"],
      "env": { "THREATCLUSTER_API_KEY": "tc_live_..." }
    }
  }
}

Cursor — one-click: see listings/cursor-deeplink.md, or add to ~/.cursor/mcp.json / .cursor/mcp.json:

{ "mcpServers": { "threatcluster": { "command": "npx", "args": ["-y", "threatcluster-mcp"], "env": { "THREATCLUSTER_API_KEY": "tc_live_..." } } } }

VS Code (.vscode/mcp.json; the input prompts for the key instead of storing it in the file)

{
  "inputs": [{ "type": "promptString", "id": "tc-key", "description": "ThreatCluster API key", "password": true }],
  "servers": {
    "threatcluster": { "type": "stdio", "command": "npx", "args": ["-y", "threatcluster-mcp"], "env": { "THREATCLUSTER_API_KEY": "${input:tc-key}" } }
  }
}

Windsurf (~/.codeium/windsurf/mcp_config.json) — same mcpServers block as Claude Desktop.

Zed (settings.json)

{ "context_servers": { "threatcluster": { "command": { "path": "npx", "args": ["-y", "threatcluster-mcp"], "env": { "THREATCLUSTER_API_KEY": "tc_live_..." } } } } }

Check a configuration without starting a client: THREATCLUSTER_API_KEY=... npx -y threatcluster-mcp --check (prints where the key came from and where it will be sent — never the key).

3. Tools

Costs are ThreatCluster API credits (free keys: 100 a day). Every result carries cost (credits this call spent), budget (remaining today, from the response headers), as_of, and url fields on every cluster, entity, victim and CVE for citation.

One prompt, threatcluster_analyst, carries the analyst rules (tools first, cite every fact with the returned url, say what period you searched, leak-site listings are claims).

Errors come back as MCP tool errors with the API's own message: 401 tells the agent to set THREATCLUSTER_API_KEY and where a free key comes from, 429 carries the Retry-After, 403 names the missing scope or the lookback window and the plan that lifts it.

Security

• The key is read from THREATCLUSTER_API_KEY (then TC_REFRESH_TOKEN, then the tc-cli store) and sent only as the X-API-Key header (or Authorization: Bearer for a JWT minted by tc login) to THREATCLUSTER_API_BASE.
• It is never logged, never printed by --check, never written to disk, and scrubbed from every error string; the test suites assert the key is absent from all stdout and stderr bytes, including on a 401 whose body quotes it.
• stdio only. No outbound connection other than the API. No analytics.
• All tools are read-only (readOnlyHint: true) and validate arguments against the spec before any request is made.

Environment

Repository layout

tools/tools.json   the single source of truth: tools, schemas, endpoint mapping, credits, prompt
python/            PyPI package (hatchling; mcp + httpx)          -> console script threatcluster-mcp
node/              npm package (TypeScript; @modelcontextprotocol/sdk + zod) -> bin threatcluster-mcp
tests/fixtures/    recorded API responses and tool outputs; both packages must replay them identically
listings/          directory manifests and submission copy (Smithery, MCP registry, Glama, mcp.so, Cursor, VS Code)
.github/workflows/ CI (both suites) and tag-triggered publishing (PyPI trusted publishing, npm provenance)

python3 tools/sync_tools.py copies the spec into both packages; CI fails if the copies drift.

Development

pip install -e "python/[test]" && (cd python && pytest)
cd node && npm install && npm test
# live validation against a real API (spends ~30 credits, re-records tests/fixtures):
THREATCLUSTER_LIVE=1 THREATCLUSTER_API_KEY=... THREATCLUSTER_API_BASE=... pytest python/tests/test_live.py

Related: the tc CLI, the API reference, the agent-tool recipe for a bare tool-calling function, and the integration guide.

GPL-3.0-or-later © ThreatCluster Ltd.