There are two AI Sync connections, and they do different jobs. Most teams want both.
docs.aisync.link/mcp is not a web page. It is a machine endpoint — open it
in a browser and you get a raw JSON error, because it only speaks to AI clients
over the MCP protocol.
This page is the documentation for it. You are in the right place.
📚 Docs MCP — reads the manual
https://docs.aisync.link/mcp
Answers how does this work from these pages. Read-only, no login, nothing to approve. Works for anyone.
🔧 Product MCP — reads your account
https://mcp.aisync.link/mcp-oauth
Answers what are my numbers and makes changes. Needs your login. See Connect MCP.
🗺️ How the two connections fit together#
Neither connection alone gives you the last box. The docs know what a falling connect rate usually means; only your account knows that yours fell.
The docs connection is the quicker win. It needs no account, so you can give it to a new rep, a contractor, or anyone helping you set up.
🔌 Connect in one command#
claude mcp add --transport http aisync-docs https://docs.aisync.link/mcpNo login step. It works immediately.
🧰 What your agent gets#
| Tool | What it does |
|---|---|
| search_docs_ai_sync | Searches every page and returns the relevant passage with a link. |
| query_docs_filesystem | Browses the docs like a filesystem, for when a question spans several pages. |
| submit_feedback | Reports a documentation problem straight to the docs team. |
That last one is worth using. If your agent gives you a wrong or confusing answer from the docs, tell it to submit feedback. The page gets fixed for everyone.
🔄 What happens when you ask#
Your agent cites the page it used. If an answer looks wrong, open the link and check it — and if the page itself is wrong, tell your agent to submit feedback.
💬 Questions it answers well#
Understanding your numbers
- "In AI Sync, what's the difference between answered and connected?"
- "Why would a connect rate drop without dials dropping?"
- "What does 'answered, no rep' mean and why does it matter?"
These pull from How calls actually work rather than from whatever the model assumes a dialer does.
Setting things up
- "How do I map dispositions to pipeline stages?"
- "How many phone numbers do I need for five reps on three lines?"
- "How do I turn on predictive pacing?"
Fixing something
- "My dashboard shows zeros but we're dialling. What's wrong?"
- "A rep shows a dash instead of a connect rate. Is that a bug?"
- "The screening message isn't playing on my phone."
Training a new rep
- "Explain what each banner colour means in the dialer."
- "What should a rep do when a call gets screened?"
Point a new rep's own assistant at the docs and let them ask instead of interrupting you.
🤝 Using both together#
With both connections live, your agent can answer a question that neither could alone:
"My connect rate dropped this week. Pull my actual numbers, then tell me what the docs say usually causes that."
The product MCP fetches your outcome breakdown. The docs MCP explains that rising Declined and Screened counts point at caller ID rather than at your reps. You get the diagnosis, not just the number.
Only the product connection can change anything. The docs connection is read-only, so there is no risk in giving it to anyone.
📄 Other ways agents read these docs#
Some tools prefer a plain text feed over MCP. Both are always current:
https://docs.aisync.link/llms.txt— an index of every pagehttps://docs.aisync.link/llms-full.txt— the full text of the whole site
Paste either URL into any assistant that accepts a link.