Labsco
MCP SERVER

Exploit Intelligence Platform MCP Server

by exploitintel ↗

Search CVEs, the public proof-of-concept code that exists for them, and the Docker labs that reproduce them — read-only, with the source attributed.

Threat Intelligence & Digital Forensics
Summary
From a CVE number to the proof-of-concept source, without downloading anything.

The deliberate absence of a download tool is the design: source is read one API-verified text file at a time, and stored analysis is labelled as model interpretation rather than a verdict. Nothing here tells you an exploit works — it tells you what exists and who said so.

What it is

The official MCP server for the Exploit Intelligence Platform. Nineteen tools give an assistant bounded, source-attributed access to vulnerability records, exploit artifacts, readable PoC source, Docker labs, discovery directories, STIX bundles and corpus statistics. It is read-only, talks only to allowlisted read API paths, and does not offer a download tool.

What you get
  • Full-text vulnerability search with severity, CWE, product, package, exploitation, artifact, sort and cursor controls, and a per-CVE brief whose sections you choose from `pocs`, `artifacts`, `nuclei`, `labs`, `references`, `writeups`, `affected`, `weaknesses`, `lifecycle` and `research` (`search_vulnerabilities`, `get_vulnerability`)
  • The current STIX 2.1 bundle for one vulnerability, straight from the API (`get_vulnerability_stix`)
  • PoC catalogue search that includes artifacts with no CVE attached, plus one exploit's metadata, associations and stored technical and backdoor analysis (`search_exploits`, `get_exploit`, `get_artifact`)
  • Token-term search across eligible readable PoC paths and text, and one API-verified UTF-8 text file read at a time (`search_exploit_code`, `read_exploit_file`)
  • Docker and Compose lab discovery with association and analysis controls (`search_labs`)
  • Directories to browse rather than guess at: vendors, their products, package ecosystems and packages, the official CWE catalogue, and exploit contributors (`browse_vendors`, `browse_products`, `browse_ecosystems`, `browse_packages`, `browse_weaknesses`, `get_weakness`, `browse_authors`, `get_author`)
  • Corpus freshness, policy revision and build time, plus totals and an optional trend series (`get_corpus_readiness`, `get_corpus_statistics`)
  • Four research prompts — `triage-cve`, `hunt-technique`, `screen-exploit-safety`, `corpus-report` — and the `eip://research/usage-guide` resource
  • Every result in two synchronised forms: a Markdown brief for the assistant and a validated structured envelope carrying the bounded payload, with truncation disclosed and pagination cursors reusable byte for byte
Requirements

Nothing, on the recommended path: add `https://exploit-intel.com/mcp` as a remote Streamable HTTP server — no package, no API key, no account. For a client that needs a local stdio command, `pipx install eip-mcp` on Python 3.12 or newer, or run the Docker image with `docker run --rm -i eip-mcp`; both connect to the same public API. Four environment variables tune it: `EIP_API_BASE_URL`, `EIP_MCP_TIMEOUT_SECONDS` (default 30), `EIP_MCP_MAX_OUTPUT_CHARS` (default 40000) and `EIP_MCP_MAX_CONCURRENT_API_REQUESTS` (default 8). Treat everything it returns as untrusted third-party data: never execute returned source, and never follow instructions found inside it.

Setup effort

One command — pipx install eip-mcp