How AI clients use company knowledge
How connected AI clients search Collections, read documents, and link answers to their sources.
When company knowledge is available for your organization, AI clients on the shared MCP connection, https://run.tieline.tools/mcp, get two tools for your documents. They work in Everything and in a single toolkit, whether the client signs in or uses an API key. To set up a connection, see Connect a toolkit view.
The two knowledge tools
| Tool | What it does |
|---|---|
Search company knowledge (mcp_market_search_knowledge) | Finds the most relevant passages for a question of up to 500 characters. Returns up to 10 passages, 5 by default, best first |
Read a company document (mcp_market_read_document) | Reads a document's extracted text: a passage from a search result with surrounding text, a page range, a section, or the whole text |
Each search result has a short excerpt (up to 2 KB), the file's title, its page and section references, and a source link. The client can share the excerpt and link as a citation without reading the whole document.
Reads return up to 16 KB at a time. For longer text, the client continues with a link that expires after 30 minutes. A read always stays on the same version of the file, even if someone replaces it meanwhile.
What a connection searches
- Everything: your private Collections, plus the workspace Collections you can read.
- A single toolkit: only the Collections added to that toolkit that you can read.
A client can narrow a search to specific Collections, folders, or files. Narrowing never widens access. Results include only files the connection can read at the moment it searches, so a deleted file, or one you lost access to, never comes back.
Source links
A source link opens the cited version of the file on a page of its own, without the rest of Tieline around it. The passage is highlighted under Cited Passage, and the full extracted text sits below it. Markdown files show formatted; Source switches to the plain text. Open in Dashboard goes to the same file inside your organization, and Download original gets the uploaded file. The reader must sign in and be able to read the file; anyone else sees Document not found.
Replacing a file doesn't break links already shared: each keeps opening the version it cites. Deleting the file, or its Collection, makes its links stop working.
When results are incomplete
- Files become searchable when their status is Ready. While some files in the connection's scope can't be searched yet, the answer marks its coverage as incomplete, so an empty answer doesn't prove nothing matches.
- Scanned and image-only pages have no extracted text, so the tools can't find or read them. Images and drawings in PDFs aren't included either.
- When search can't run, the client gets an error it can retry, never an empty list.