Upside MCP Usage

What is the Upside MCP?

The Upside MCP gives AI assistants direct access to your Upside instance. Rather than logging into the dashboard and clicking through reports, you can simply ask your AI tool a question and it will query your data, analyze it, and return the answer.

🔒

Security note: The MCP is read-only. It cannot modify, delete, or write anything to your upstream systems. When you query through the MCP, the AI accesses data that is already within Upside's secure environment.

---

Getting Started

Before you begin, follow the instructions in the Upside MCP Setup Guide to connect your preferred platform. You may need a system admin to enable your use of the MCP.

---

Activating Upside in Your AI Tool

Once your admin has completed the setup, each team member needs to activate the Upside connection using their own Upside credentials. This is a one-time step.

Claude

1. Go to claude.ai/settings/connectors
2. Find the Upside connector and toggle it on
3. Start a new conversation and include "Using Upside, …" in your prompt

ChatGPT

1. Go to chatgpt.com/apps?q=upside
2. Click the Upside connector and enable it
3. Start a new conversation and @mention the Upside connector in your prompt

Claude Code / Cursor / Codex

These tools connect via MCP configuration that your admin sets up at the project or system level. If you don't see the Upside tools available, check with your admin that the MCP config has been applied to your environment.

---

How to Get the Best Results

The key to getting great answers is being specific. The more context you give, the better the response.

How to initiate the Upside MCP

Each platform has their own way of directing the AI to connect to and use the Upside MCP, rather than trying to answer from general knowledge/other data sources. see the table below on how to invoke the MCP.

Share Your Intent, Not Just Instructions

The best prompts don't just say what to do, they explain why. This is often called "commander's intent": give people your goal and reasoning so they can adapt when things don't go exactly as expected. The same principle works with AI.

When you tell the AI the bigger picture — what decision you're trying to make, what hypothesis you're testing, or what presentation you're building — it can make smarter choices about how to query the data, what to highlight, and what follow-up insights to surface on its own.

❌

Just instructions

"Show me a table of deals that closed last quarter with their marketing touches."

✅

With intent

"I'm preparing for a board meeting where I need to justify increasing our marketing budget, and discuss investing in expanding our webinar program. Using Upside, analyze deals that closed last quarter and show me how webinar touches correlated with win rates compared to other channels. Highlight any patterns that you observe with our current program.”

Be Specific, Not Vague

The AI works best when you tell it exactly which subset(s) of data to look at and what kind of analysis you want.

❌

Too vague

"Tell me what's working."

✅

Much better

"Using Upside, look at deals that closed in Q4 2025 with ARR above $50K. What marketing channels were present before the deal moved to Stage 3? Summarize the common patterns."

When the AI understands your goal, it goes beyond pulling raw data — it structures the output to actually answer your question.

Types of Questions You Can Ask

Quantitative: finding patterns in numbers

• "What percentage of deals that had a webinar touch point before Stage 2 ended up closing, versus deals that didn't?"
• "Show me all campaigns that created opportunities post-engagement in the last 6 months, sorted by pipeline created."
• "How many deals closed with a personal email associated with a member of the buying group?"
• "What is the average deal cycle for deals where marketing was involved versus deals where it wasn't?"

Qualitative: going deep on specific deals

• "Go deep into the Acme Corp deal and figure out exactly what happened. What touchpoints occurred, what did people say on calls, and why did we win?"
• "Look at our last 5 lost enterprise deals. Read through the call transcripts and emails. What are the common reasons we're losing?"
• "For the deals that moved from Stage 3 to Stage 4 this quarter, analyze what content or touchpoints happened right before that transition."

Account research

• "Summarize everything we know about [Account Name]: all touchpoints, emails, meetings, and buying group members."
• "How have we interacted with [Person Name] over the last year? What meetings did they attend, what emails were exchanged?"

Channel and campaign analysis

• "What channels are most present in our closed-won enterprise deals? Break it down by awareness, activation, and closing stage."
• "Compare the performance of our hosted webinars versus third-party webinars in influencing pipeline."
• "Look at our referral program. Who are our top referrers and how did they engage with us before making introductions?"

Custom cohort analysis

• "Create two cohorts: deals where the AE had 3+ contacts engaged versus fewer than 3. What's the close rate difference?"
• "Find deals where we had a webinar, then at least 3 follow-up touches before closing. What was the average time between the webinar and close?"
• "Are there any AEs who consistently use the same rejection reason regardless of what happened on calls?"

Tips for Better Prompts

1. Specify a time range. "In the last quarter" or "deals that opened in 2025" helps narrow results.
2. Name the fields you care about. If you have specific Salesforce fields (like opportunity type or record type), mention them.
3. Ask for the format you want. "Show this as a table" or "Summarize in 3 bullet points" or "Build me a mini-app for this."
4. Iterate. Ask follow-up questions in the same conversation. The AI remembers context: "Now break that down by sales rep" or "Remove renewals from that analysis."
5. Don't worry about SQL. You never need to write queries. Just describe what you want in plain English and the AI figures out how to pull it.

---

Advanced: Deep Research with Subagents

Some AI platforms (like Claude Code and Craft Agents) support subagents — the ability to spin up multiple parallel analyses at once. This is like having a team of analysts each investigate a different deal or segment simultaneously, then report back with findings.

This is especially powerful for:

• Multi-deal deep dives: "Analyze each of our 10 largest lost deals from last quarter in parallel. For each one, read through the emails and call transcripts and summarize why we lost."
• Comparative analysis: "Look at our top 5 performing AEs and our bottom 5. In parallel, analyze their deal patterns, touchpoint cadence, and multi-threading behavior. Then compare and tell me what the top performers do differently."
• Segment exploration: "Split our closed-won deals into SMB, Mid-Market, and Enterprise cohorts. Analyze each segment's buying journey in parallel and build a comparison."

💡

How to trigger it: Ask the AI to analyze multiple things "in parallel" or "simultaneously." On supported platforms, it will automatically delegate to subagents. On platforms that don't support subagents, the AI will still run the analyses — just sequentially.

---

Building Mini-Apps

Modern AI tools can do more than answer questions — they can build fully functional, interactive software on the fly. We call these mini-apps: lightweight, self-contained dashboards and tools that the AI generates from your data in response to a single prompt. Think of them as custom reports that you can click through, filter, and explore — without any design or engineering work. They're one of the most powerful ways to get value from the Upside MCP.

What Are Mini-Apps?

Mini-apps are self-contained visual reports that look and feel like a professional dashboard. They include charts, tables, filters, and interactive elements, all generated from a simple prompt.

How to Create One

After getting results from a query, simply ask:

💡

"Build me a mini-app that visualizes this data with filters for time period and deal size."

The AI will read the Upside documentation on mini-app formatting and generate a complete, styled HTML file.

Two Types of Mini-Apps

Static mini-apps (you can build these yourself):

• Created locally on your machine using any AI tool connected to Upside
• Data is baked into the app at creation time
• Great for point-in-time analyses, presentations, and sharing snapshots
• Open the HTML file in any browser to view

Live mini-apps (deployed by the Upside team):

• Hosted in your Upside dashboard under the Apps section
• Pull live data every time they're opened, so they're always up to date
• Built by your Upside team or requested through your account manager
• Accessible to everyone on your team with Upside access

Real Examples of Mini-Apps Customers Have Built

• Channel Attribution Matrix: shows which channels were present in each closed deal, with quotes from calls and emails as evidence
• Campaign Follow-Up Tracker: interactive table of campaigns with post-engagement opportunity metrics and filters
• Referral Network Map: who your top referrers are, how you engaged them, and which deals they influenced
• Lost Deal Analysis: AI-generated reasons for losing deals pulled from call transcripts and emails, with recommendations
• Sales Multi-Threading Report: which deals have fewer than 3 contacts engaged, broken down by team and manager
• Deal Engagement Timeline: visual timeline of all touchpoints across a deal's lifecycle, by channel

Getting a Mini-App into Your Dashboard

If you've built a static mini-app that you'd like everyone on your team to access:

1. Share the HTML file with your Upside account team
2. We'll deploy it to your Upside dashboard under the Apps section
3. Your whole team can access it from there

---

What Data is Available?

Your Upside instance contains a unified, cleaned version of your go-to-market data:

• Accounts: company records from your CRM
• People: contacts with deduplicated identities across systems
• Buying groups: who is actually involved in each deal
• Opportunities: deal records with stage history
• Emails: individual messages (not raw Salesforce activity blobs) with proper timestamps
• Meeting transcripts: what was said on calls (if Gong or similar is connected)
• Campaign memberships: who engaged with which campaigns
• Web activity: page visits and engagement signals
• Touchpoints: every interaction, categorized by channel and type

The AI understands how to navigate all of these tables and join them together. You just ask the question.

---

Frequently Asked Questions

Do I need to know SQL or how to code?

No. You just describe what you want in plain English. The AI writes the queries for you.

Can the MCP change or break anything in my systems?

No. It's completely read-only. It cannot modify any data in Upside, Salesforce, HubSpot, or anywhere else.

Which AI model should I use?

For Claude, use Opus as it's significantly smarter and gives better results. For ChatGPT, use GPT-5.2 or later. For Cursor, we recommend Claude Opus as the model.

Why does the first query take a little longer?

On the first query in a new conversation, the AI explores the data schema to understand how your tables are structured. After that, follow-up queries in the same conversation are much faster.

Can I share what the AI builds with my team?

Yes. Static mini-apps are just HTML files, so share them however you like. For dashboards that need live data and team-wide access, ask your Upside team to deploy them.

---

Need Help?

• Slack/Teams: Reach out to the Upside team in your shared channel
• Email: Contact your Upside account manager
• Office hours: Ask your account team about scheduling a live walkthrough