Assembled MCP: Customer Getting Started Guide

In this article

• What data can I access through the MCP?
• How do I set up the MCP?
• How do I write good prompts?
• What prompts should I start with?
• How do I use the MCP to compare forecast vs. actual?
• What are common questions and answers?

The Assembled MCP (Model Context Protocol) server connects your live Assembled data directly to AI tools like Claude, ChatGPT, and others. Once connected, you can ask questions about your support operations in plain English. There's no manual report-building, no exporting to spreadsheets, and no waiting on an analyst.

This article covers what the MCP can do, how to set it up, and how to get the most out of it once you're connected.

What data can I access through the MCP?

The MCP gives you read and write access to your core operational data in Assembled.

Available for read (query and analyze):

• SLA, AHT, adherence, volume, and all core performance metrics
• Real-time intraday data
• Schedules and coverage
• Workforce composition (agent skills, team, site, and queue assignments)
• Forecast data and historical accuracy
• Labor law compliance (breaks, lunches, time off)

Available for write (take action):

• Adjust forecasts and mark outliers
• Publish overtime (OT) and voluntary time off (VTO) slots

Not available today:

• AI support agent data (chat, email, and voice AI resolution metrics). This is outside the MCP's scope, which is built on the workforce management (WFM) metrics platform.
• Staffing metrics. There's no defined timeline for this currently.
• Pending PTO requests. You can query total PTO hours requested per agent over a time frame, and approved PTO, but not pending dates or request types.
• People-page or headcount data. Agent skill timelines and queue assignments from the people page aren't yet surfaced.
• Some AI metrics. These are still rolling out as part of the broader metrics platform migration.

Historical data is available from October 2025 onward.

How do I set up the MCP?

The MCP server URL for all clients is:

https://data.mcp.assembledhq.com/mcp

For step-by-step setup instructions, see the guide for your client:

• Set up the Assembled MCP in Claude.ai
• Set up the Assembled MCP in ChatGPT
• Set up the Assembled MCP in Claude Desktop
• Set up the Assembled MCP in Goose
• Set up the Assembled MCP in Dust

How do I write good prompts?

You don't need to know metric names or filter syntax. Describe what you want to know in plain English. The more context you give, the more precise the answer.

Tips for better results:

• Use specific dates, not relative terms. "The week of June 9" or "June 1-14" returns more reliable results than "last week."
• Specify channel or queue when relevant. "On chat" or "for the phone queue" gets sharper answers.
• Ask follow-up questions. The AI holds context, so you can drill deeper from the first answer.
• Tell it which tool to use if needed. If it's not pulling from Assembled automatically, add "Use Assembled" to your prompt.
• Start broad, then narrow. "What metrics can I analyze?" is a good first prompt before diving into specifics.
• If a query returns no data, the metric may not be available yet, or the date range may be outside the October 2025 window. Try rephrasing or adjusting your timeframe.

What prompts should I start with?

Discover what's available:

• "What metrics do I have access to in Assembled?"
• "What data can I query through this connection?"

For WFM teams

Diagnose SLA misses:

• "Where and why did we miss SLA the week of June 9? Break it down by interval and tell me whether it was shrinkage, AHT, or volume."
• "Which queues or channels missed SLA most often in May?"

Inspect schedules and coverage:

• "Show agent count by site, channel, and skill right now. Flag anywhere we're below scheduled headcount."
• "Show me who's scheduled on chat tomorrow between 10am and 2pm ET."
• "How many manual schedule changes did BPO managers make between June 1-14, and how many were reverted?"

Analyze forecasts:

• "What's the forecast for email volume next week by day?"
• "How accurate was the chat forecast for May vs. actuals, broken down by interval?"
• "If volume jumps 25% next week, where do we go into deficit, and what reps per hour (RPH) do we need to still hit SLA?"

Take action (write):

• "Apply a +40% volume override to chat for the NFL Sunday window (Sun 1pm to 11pm ET) across all US queues."
• "Publish OT on phone for every interval next week where forecast exceeds scheduled coverage by more than 10%."
• "Generate VTO offers for any 30-minute interval this week where we're more than 15% overstaffed on email."

For support leadership

Daily pulse:

• "Show me yesterday's volume by queue across all sites."
• "Did we hit SLA across all queues yesterday?"

Weekly and monthly reporting:

• "Summarize SLA performance for the week of June 9 for the leadership update."
• "Show SLA, AHT, and adherence for each team this week, ranked."
• "Which BPO sites are underperforming so far in Q2?"

Planning and what-if scenarios:

• "If we cut headcount 10%, what's the SLA impact?"
• "Bump email forecast 25% for Black Friday, midnight to midnight PT on Nov 28."

How do I use the MCP to compare forecast vs. actual?

One of the most useful things you can do with the Assembled MCP is compare what your team was planned to handle against what actually happened. This helps you understand whether service gaps came from bad forecasting, staffing execution issues, or unexpected demand. That way, you can make smarter decisions going forward.

Every week, Assembled generates a forecast: expected contact volume, handle time, required agents, and whether scheduled capacity covers demand. Forecast vs. actual compares those predictions to what really happened, bucketed across your time window.

What you can query:

• Volume: actual contacts received vs. forecasted, and where volume was above or below plan.
• Handle time: whether cases took longer or shorter than forecasted.
• Staffing: agents required vs. scheduled, and net staffing position (over or understaffed, and by how much).
• Service level: whether actual SLA performance aligned with what the forecast model predicted.

How results are bucketed by time window:

Important parameters:

• Channel is required. You must specify email, chat, or phone. An aggregate "all channels" view isn't supported. If you're not sure what channels your environment uses, just ask and the MCP will look it up first.
• Queue is optional. Leave it blank to aggregate across all queues, or specify a queue by name.
• Time range limit. This works best for windows up to approximately 6 months. For longer historical analysis, use the metrics query tools instead.

Example prompts:

• "How did actual email volume compare to forecast last week?"
• "Were we understaffed or overstaffed on email last Monday?"
• "Show me the staffing gap for our chat queue over the past 2 weeks. Where were we most short?"
• "How accurate was the chat forecast for May vs. actuals, broken down by interval?"

How to interpret results

A staffing gap doesn't always mean a forecasting problem. If volume was accurately forecasted but you were still understaffed, the issue may be in scheduling or execution. Use adherence and conformance metrics to investigate further.

A volume miss doesn't always mean bad forecasting. Unexpected events like product launches, outages, or campaigns can drive volume no model would have predicted.

Overstaffing is also worth tracking. Persistent overstaffing can indicate forecast inflation or scheduling inefficiency.

Results always reflect your company's master schedule as the staffing baseline, not draft or alternative schedules.

What are common questions and answers?

I'm not an admin. Can I still use the MCP?

Not yet. MCP access is currently limited to admin users. Ask your Assembled admin to run queries on your behalf or request a role update. Role-based access for non-admins is on the roadmap.

My company uses enterprise ChatGPT and IT won't allow custom connectors. What do I do?

Assembled has submitted for ChatGPT marketplace review, but approval timelines are outside our control. Reach out to your CSM. We can provide security documentation to support your IT team's review.

I set it up but it's not pulling from Assembled. What's wrong?

Try adding "Use Assembled" explicitly to your prompt. If that doesn't work, check the following:

1. You're logged into Assembled in the same browser session.
2. Your account has admin-level MCP access enabled.
3. Tool permissions are set to Needs approval or Always allow, not left at default.

Does my query return data for my entire company or just me?

Your MCP connection is scoped to your Assembled instance and returns data for your organization, tied to your admin login.

How far back does data go?

Historical data is available from October 2025 onward.

My query returned nothing. Is the data missing?

A few possibilities: the metric isn't available yet, the date range may be before October 2025, or the phrasing may need to be more specific. Try asking "What metrics do I have access to in Assembled?" to see what's currently queryable.

Can I query data from multiple Assembled instances at once?

Not currently. The MCP server is scoped to one company instance at a time based on your login. To point the connection at a different instance, disconnect the MCP server, then reconnect and log in with the account tied to that instance.

Is my Assembled data being used to train AI models?

No. Assembled doesn't use your data to train any models. The AI model runs on the client side (Claude or ChatGPT), not within Assembled. Model behavior is governed by your chosen AI provider's policies.

Does Assembled log what I ask?

Assembled captures standard operational logs of MCP activity (which tool was called, timing, success or failure) to support troubleshooting. Assembled doesn't capture your natural-language prompts or questions.

What compliance certifications does Assembled hold?

Assembled is SOC 2 Type II, GDPR, CCPA, and HIPAA compliant. Data is encrypted in transit and at rest. See trust.assembled.com for full details.

What's the difference between the Assembled MCP server and MCP connectors in Assembled AI?

These are two different things. The Assembled MCP server (this guide) is you connecting your LLM to Assembled's data to ask questions about your WFM metrics. MCP connectors in Assembled AI are Assembled's AI agents connecting outbound to your tools (ticketing system, CRM) to take actions during customer interactions.