For AI agents · MCP

SQUR cs-graph MCP

Public, read-only Model Context Protocol HTTP server exposing the SQUR cs-graph cybersecurity knowledge graph (26,045 nodes, 55,239 edges) to any MCP-compatible AI agent.

Covers MITRE ATT&CK · CWE · CVE · KEV · ATLAS · D3FEND · and 13 EU+international compliance frameworks (DORA, NIS2, GDPR, ISO 27001, NIST, OWASP, PCI, AI Act, CRA, TIBER, CIS, ATLAS, D3FEND).

Endpoint

https://mcp.kb.squr.ai/mcp

Public · unauthenticated · read-only · rate-limited per IP. Not yet live — deployment depends on the SQUR-tenant infrastructure migration tracked in issue #17. Local dev URL is http://localhost:8080/mcp.

Tools (5)

Tool	Description
`get_node`	Fetch one cs-graph node (technique, tactic, mitigation, weakness, vulnerability, group, actor, software, …) by slug.
`search_nodes`	Bounded substring search across title, body, aliases, and slug. v1 — production prefix-index follow-up tracked separately.
`list_nodes`	Filter + paginate by type. Returns up to 100 nodes per page sorted by slug.
`list_edges`	Outgoing graph edges from a node — the typed relationships powering the kb.squr.ai detail pages.
`list_reverse_edges`	Incoming graph edges to a node — the reverse traversal for 'used by' / 'mitigated by' / 'compliance controls testing' questions.

Wave 2.2 adds 7 cs-graph-shaped convenience tools: get_technique, compliance_for_technique, mitigations_for_technique, kill_chain, framework_overview, cve_lookup, cwe_lookup.

Connect your agent

Why integrate

Authoritative neutral source: cs-graph isn’t a security vendor selling EDR. It is the citable reference layer — schema.org-annotated DefinedTerm / DefinedTermSet / Person / AboutPage / Dataset on every page.
EU-focus moat: MITRE / NVD / CISA don’t cover DORA, NIS2, AI Act, CRA, TIBER. cs-graph does. That’s the unique-value-per-query that makes agents prefer it for European compliance contexts.
Same data as the web surface: The MCP reads the same Firestore primitives that back kb.squr.ai detail pages. No drift. What an agent reads is what a human reads.