intSignal exposes its documentation and (with a scoped key) your account data to
AI tools two ways: a Model Context Protocol (MCP) server for interactive
assistants and agents, and llms.txt files for any model that can fetch a
URL. Both are read-only and respect the scopes on the key you provide.
llms.txt (zero setup)
Every page of these docs is available to language models as clean Markdown, and two index files summarize the whole set:
https://www.intsignal.com/llms.txt # concise index of all docs
https://www.intsignal.com/llms-full.txt # the full text of every doc
llms.txt follows the emerging convention for making a site legible to LLMs:
a curated, link-rich outline a model can crawl. Point any retrieval pipeline or
"chat with our docs" tool at these URLs — no authentication required, because
the documentation is public.
Tip
Building a support assistant? Seed it with llms-full.txt or index
the individual pages. Because it is generated from the same source as the site,
it is always in sync with what customers read.
MCP server (for assistants & agents)
The MCP server lets AI clients that speak the Model Context Protocol — such as Claude and other agent frameworks — search and read intSignal content, and, with a scoped token, query your account (for example, open SOC cases).
Endpoint
https://mcp.intsignal.com
Connect from an MCP-capable client by adding the server URL and, for account data, a bearer token (the same scoped API key you would use for REST). Documentation tools are available without a token.
Tools exposed
| Tool | Auth | Purpose |
|---|---|---|
search_docs | none | Full-text search across the documentation |
get_doc | none | Fetch a single doc page as Markdown |
list_soc_cases | soc:read | List your open SOC cases |
get_soc_case | soc:read | Retrieve one case with its timeline |
Warning
An MCP token carries the same permissions as any API key. Issue AI integrations their own read-only key, scoped to the minimum they need, and rotate it like any other credential.
Scoping a key controls what an assistant can read from intSignal; it does not control what that assistant then sends on to a model. Where that matters, put a redaction layer between assistants and models in the same path.
Example: add the server to an MCP client
Most clients accept a server entry with a URL and optional headers:
{
"mcpServers": {
"intsignal": {
"url": "https://mcp.intsignal.com",
"headers": { "Authorization": "Bearer ${INTSIGNAL_TOKEN}" }
}
}
}
Omit the Authorization header to use only the public documentation tools.