How MoltSets' AI Support Agent Resolves 90% of Tickets

At MoltSets, our AI support agent answers roughly 90% of customer tickets without a human (me) touching them. That number gets asked about a lot, usually with the assumption that it comes down to model choice or prompt engineering. It doesn't. It comes down to what the agent has access to.

Three things make that possible: documentation the AI can actually rely on, tools that put customers directly into the right workflow, and a live view into each customer's account. Here's how each one works.

1. Documentation built for both humans and AI

Our API and MCP documentation lives on Mintlify, and it's written to be genuinely thorough, not just a reference for parameters and endpoints. It covers the reasoning behind how MoltSets works, the edge cases customers run into, and the context needed to answer "why" questions, not just "how" ones.

The part that matters most for support is that this documentation is external and kept current with every product change. When we ship an update, the docs update with it. That means the AI agent is always pulling from the same source of truth a developer would use, not a stale internal wiki or a support macro that someone forgot to revise six months ago. Most AI support failures come from the agent confidently repeating outdated information. Ours doesn't have that problem, because there isn't an outdated version floating around for it to find.

2. A Skill Library that gives customers the right tool for the job

Documentation answers questions. It doesn't get work done. That's what our Skills Library is for.

We built a set of pre-built Claude Skills covering the workflows customers actually run: HubSpot and Attio MCP integrations, LinkedIn enrichment, SEO audits, and more. When a support conversation reveals that a customer wants to do something rather than understand something, the agent can point them straight at the skill that does it, instead of trying to talk them through a multi-step process in a chat window.

This matters because a lot of support tickets aren't really support tickets. They're customers trying to accomplish a task and hitting friction along the way. Giving the agent (and the customer) a direct tool for that task removes the friction rather than explaining around it.

3. Real-time account access

This is the piece that turns the agent from a knowledgeable assistant into one that can actually resolve tickets: it has live, read access to the customer's own account data through a custom API.

When a customer asks why their search is being rate-limited, or how many tokens they have left, the agent isn't guessing or asking the customer to go check a dashboard. It's looking at data like this, pulled in real time:

{
  "id": "XXXXXXXXX",
  "name": "Example Co",
  "status": "active",
  "plan": "free",
  "token_balance": 993,
  "unlimited": false,
  "fair_use": {
    "enrich": { "requests": { "5h": { "limit": 5000, "used": 0, "remaining": 5000 } } },
    "search": { "requests": { "5h": { "limit": 200, "used": 0, "remaining": 200 } } }
  },
  "search_trial": { "results_total": 100, "results_used": 31, "results_remaining": 69, "max_per_call": 5 },
  "external_token_balance": 100,
  "external_token_allowance": 100,
  "api_key_count": 0,
  "user_count": 1
}

From this, the agent can answer plan status, remaining token balance, rate limit headroom, and trial usage without escalating. Most account-related tickets stop being tickets at all, because the question gets answered with the customer's actual numbers instead of a generic policy explanation.

Why this combination works

None of these three pieces would get you to 90% on their own. Good documentation without account access still means the agent gives a correct but generic answer to a specific question. Account access without documentation means the agent can see the data but can't explain what it means or what to do about it. Skills without either just gives customers a faster way to hit the same wall they were already stuck at.

Put together, the agent can explain how something works, show the customer exactly where they stand in their own account, and hand them the tool to fix it, all in the same conversation. That's the difference between an AI that answers questions and one that resolves tickets.