The docs for agents
Every page of this documentation has a markdown copy, and two index files list them all, so a model can read the docs the way you do. This page says where they are and what they leave out on purpose.
Updated October 19, 2026Markdown
This documentation is written for people, and it is also served in a form a model reads without scraping. Point an assistant at it when you want it to answer a question about sourcesignal, to draft a note for a teammate from the docs, or to work inside sourcesignal for you over MCP and understand the screens you see. Nothing on these addresses differs from the pages themselves: the same articles, the same words, the same limits.
Where the docs are, for a model
| Address | What it holds |
|---|---|
| llms.txt | The index: one line per article with its markdown address and its summary, then the main pages of the site |
| llms-full.txt | Every article in one file, in reading order |
Any article's address with .md appended | That article as markdown, with its section, the day it was updated and its address for people |
Each article's page also names its markdown copy in its head, as an alternate of type text/markdown, so a client that asks for markdown finds it without guessing the address.
- The index is the place to start. It is short, and each line carries the article's summary, so a model picks the right article before reading it.
- Links inside the markdown copies are absolute, and a link to another article points at that article's markdown copy. A model follows them as they are.
- When you hand a person a link, hand them the page, not the markdown copy. The copy's first lines name the page's address for that.
Ask an assistant about sourcesignal
Paste the index address and your question. For example: "Read https://www.sourcesignal.ai/llms.txt, then tell me how a page gets from Opportunities into Outreach, with the article links." The assistant reads the index, picks the articles and answers with links you can open.
Connect an agent to your account
The docs explain the screens. An agent that will work inside sourcesignal for you needs the connection contract as well: the endpoint, the token, the rules and every tool. That contract is the brief on outreach for agents, served as markdown too, and the one line to paste into your agent is on that page. Connect your own agent over MCP walks you through making the token, Agent scopes and limits says what each scope allows, How the door answers says what your agent meets at each limit, and the Tool reference lists every tool by scope.
Once connected, your agent needs no web access for the docs. The same text is served to it as resources: the index as sourcesignal://docs and every article as sourcesignal://docs/{section}/{slug}, the section and slug being the two parts of the article's address. A client that lists resources shows them; one that does not can still read the addresses above.
What the docs leave out on purpose
The documentation says what you see and what you can do. It never says how a verdict is computed: how a page earns its priority, how a mention is read as lead, list or footnote, how a page type or an outreach approach is chosen, or how a win is checked. Those methods are sourcesignal's own and are not published. Inside the app, "Why this score" and "Why this card" name the reasons for one page. The method behind them stays out of the docs and out of these files.
A model asked how any of this works should say that the method is not published and point to what is: the screens, the statuses and the limits in these articles. An answer that fills the gap with a guess is wrong, however plausible it sounds.
When the docs and the product disagree
The product changes, and every article says the day it was last updated. When a model reads a label here that the screen no longer shows, the screen is right. Tell us through Support, naming the article, and the next revision carries the fix.