Tips and Tricks when using the MCP

The Problem: Running analyses in standalone chats leads to:

• Chats maxing out faster than expected

• Loss of context when starting new conversations

• Uncertainty about whether previous work is being considered

• Frequent errors and hallucinations

• Time wasted double-checking numbers

The Solution: Create a Project from day one for any multi-step analysis or ongoing work. It will give you:

• Persistent Context: Your rules, instructions, and context carry through all chats within the project

• Consistency: Reduces errors and hallucinations by maintaining clear parameters

• Efficiency: No need to re-explain your data structure or requirements in each new chat

• Confidence: Numbers and analyses remain consistent across conversations

• Organization: All related work stays in one place

Here’s an example of how to create a project: Executive Summary Prompt

For example in the conversation below, Claude assumed that the date range should be January 1 to January 14th 2025.

By checking the generate_report tool, we could tell that the date range used shouldn’t be this one. Same goes for the metrics and dimensions it’s using to generate your answers.

​

​

Looking at the tools available in this conversation, for example, the Polar:generate_report tool has a description explaining it generates comprehensive analytics reports and specifies exactly what inputs are required (metrics, dimensions, date ranges, etc.) and what output format to expect (JSON report data).

Create a project with instructions, and attach a Brand Guidelines text file.

# Brand guidelines



Logo: bear-logo-svg.txt (SVG inside a text file)

- Display at minimum 60px height for optimal readability

- Maintain aspect ratio when scaling



Primary text: #0F3328

Secondary text: #446058

Accent color: #00A574

Background color: #FFFFFF



Use a light background, very few boxes or borders, just dividers or shaded background where absolutely necessary.



Body font: Source Sans 3 Light 300

Heading font: Source Serif 4 SemiBold 600

Both fonts available from Google Fonts

For your logo, attach an SVG with a .txt extension. If you upload a PNG or JPG it will “see” the image and try to recreate it instead.

Result:

• Run an analysis of net sales by Top 5 Cities (using Billing Address), ranking by year from 2020 to 2025 full year

• What is the ad spend for the UK region in August 2025?
  
  ​If you’ve created a “UK” view in Polar, it will find this and use it as a filter.

• Can you give the EU Total Sales for last month?

• Can you look at the [dashboard name > report name] report and give me a summary?
  
  ​Replace the dashboard and report name with something you’ve built in Polar.

• Is this just ChatGPT on top of APIs?

  No. Claude & ChatGPT connect to Polar’s semantic layer, not raw endpoints. That’s why answers are consistent, accurate, and repeatable.

• For users that have access to multiple tenants via Workspaces, what are the implications for their MCP usage? Will all users on both accounts be able to see all of his recent chats?
  
  MCP (e.g. with Claude) - Claude queries Polar in a read-only manner. No trace of your chats with Claude are left inside Polar. Claude chats are only visible to others if you share them manually via Claude's sharing features.

• How do you make sure the AI uses the right metrics or date ranges?

  The MCP guides Claude to confirm metrics, filters, and date ranges before analysis. You’ll see it validate context live in your chat.

  Always double check the tool calls that the LLM is making. For example in the conversation below, Claude assumed that the date range should be January 1 to January 14th 2025.

  
  
  

• Will it work with my custom definitions?

  Yes. Your custom metrics, dimensions and views are being fed into Claude via the tools

• Is my data secure?

  The security of your data depends on your Claude subscription type:

  • Commercial Products (Claude for Work - Team/Enterprise, API)

    • ✅ No training on your data by default - Anthropic does not train models on data from commercial products

    • ✅ Your inputs and outputs remain private unless you explicitly opt-in to development programs

    • ⚠️ Exceptions: Data may be reviewed if you provide feedback or if conversations are flagged for policy violations

  • Consumer Products (Claude Free, Claude Pro)

    • ⚠️ Different privacy policies apply - Claude Pro is classified as a consumer product, not commercial

    • ⚠️ Consumer products follow different data usage practices than commercial offerings

    • 📝 Check your Privacy Settings to confirm your data usage preferences

  Recommendations

  1. For sensitive customer data: Consider using Claude for Work (Team/Enterprise) plans for commercial-grade privacy protections

  2. Review privacy settings: Check your current settings at Anthropic's Privacy Policy

  3. Isolate sensitive data: Take precautions when connecting Claude to services with highly sensitive information

  4. Stay informed: Privacy policies and data handling practices may evolve, so review them periodically

  TLDR: If you're handling sensitive customer data through MCP tools, a commercial Claude plan provides the strongest privacy guarantees with no model training on your data by default.
  
  ​