MCP
API vs MCP
The API gives you the raw model. The MCP server gives that model CKE 2.0 and the research tools. Connect both.
NotHoly ships two surfaces, and they do different jobs.
- The API at
https://api.notholy.com/v1is raw inference. You send messages, you get the model's answer. It is OpenAI and Anthropic compatible, so any client that speaks those APIs speaks NotHoly. The model has no retrieval and no tools of its own here. - The MCP server at
https://api.notholy.com/mcpis the intelligence layer. It gives a model Cybersecurity Knowledge Engine 2.0 (CKE 2.0) and the research tools: CVE records, package advisories, web and page reading, file analysis, passive network intel, and the skills registry. These are the same tools the research chat runs.
If you connect only the API, you get the model alone. CKE 2.0 and the tools live behind the MCP server. To get them, connect the MCP.
What each one gives you
API /v1 | MCP /mcp | |
|---|---|---|
| Inference (chat, completions, Anthropic messages) | Yes | No, it is not an inference endpoint |
| OpenAI and Anthropic compatible | Yes | — |
CKE 2.0 knowledge base (kb_search, kb_read) | No | Yes |
| CVE records and search | No | Yes |
| Package advisories (OSV) | No | Yes |
| Web search and page reading | No | Yes |
| File static analysis | No | Yes |
| Passive network intel (IP, DNS, WHOIS, ASN, site, breach) | No | Yes |
| Dark web index search | No | Yes |
| Skills registry | No | Yes |
| Transport | HTTP (REST, SSE for streams) | Streamable HTTP (MCP, JSON-RPC) |
| Auth | Authorization: Bearer or x-api-key | Authorization: Bearer |
The raw model endpoints, POST /v1/chat/completions, /v1/responses and /v1/messages, carry none of these tools. The model gets only the function tools you define and send, and your own code runs them. The MCP server is how you hand a model the NotHoly tools without writing that plumbing yourself.
Want CKE 2.0 and the tools? Connect the MCP
Point your MCP client at https://api.notholy.com/mcp and send your API key in the Authorization: Bearer header. The same key works for both the API and the MCP server.
claude mcp add --transport http notholy https://api.notholy.com/mcp --header "Authorization: Bearer $NOTHOLY_API_KEY"The end-to-end guide walks through getting a key, calling /v1 for inference, connecting the MCP in the common clients, and using each tool with example input and output.
Which path do I need?
- You only want model answers from your own code or client: use the API. Nothing else to set up.
- You want a model (in Claude Code, Cursor, your own agent) to look up CVEs, search CKE 2.0, read pages or analyze files: connect the MCP server.
- You want this data in your own code without an agent, as plain HTTP: the Research API exposes CKE 2.0, CVE, advisory, web and file routes under
/v1.