Guide

    How to build data-rich slides with Claude Code/Cowork

    AI-generated board deck with 15 slides: revenue, ARR, pipeline, product overview, client deep dives, marketing analytics, cash position, and more

    A real board deck built by Claude in 20 minutes, pulling live data from a warehouse, CRM, and meeting notes

    Connect your data to Claude with an MCP connector, add a skill file, and generate board decks, QBRs, renewals decks, and investor updates from live company data. No copy-pasting from dashboards. No hallucinated numbers.

    Board Decks
    Quarterly updates with live revenue, pipeline, product, and runway data
    QBRs
    Client-facing quarterly reviews with usage metrics and ROI analysis
    Renewals Decks
    Renewal presentations with account health, adoption trends, and value delivered
    Investor Updates
    Concise progress reports with ARR, burn, and milestone tracking

    Claude built our board deck by itself, in 20 minutes. Accurate data, accurate deal pipeline, accurate key takeaways. Branding? Not perfect, but good effort.

    I didn't edit a single thing, except pasting in two redacted screenshots. One VC at the meeting called it "better than most of our Series A portfolio companies."

    So how did we make sure it doesn't hallucinate data or produce slop? We gave Cowork three key tools: the PowerPoint connector to build slides, the Rig MCP to pull contextualised internal data, and a design skill to keep it on brand.

    Setup took 10 minutes, execution took 10 minutes. There are a few tools you can use if you're not into PowerPoint (Canva, Gamma, or Figma via their MCPs), but the data integration and skill setup is the key.

    TM
    Toby Mather
    CEO, Rig

    How to connect your data to Claude and get started

    You need three things: Claude (via Claude Code or Cowork), a data connector so Claude can access your company metrics, and the skill file that teaches it how to build decks.

    1

    Get Claude Code or Cowork

    The skill file works with any Claude interface that supports skills and MCP connectors.

    Claude Code (CLI)

    Install via npm install -g @anthropic-ai/claude-code and run claude in your terminal.

    Best for technical users comfortable with the command line.

    Claude Cowork

    A visual workspace at claude.ai/code with the same capabilities, no terminal required.

    Great for non-technical team members building decks.

    2

    Connect your data to Claude via MCP

    Claude needs access to your company data to pull real numbers into the deck. You do this through MCP connectors. There are a few approaches, each with different trade-offs.

    Rig (recommended)Best results

    Rig syncs your warehouse, CRM, and analytics into a single SQL-queryable connector with semantic context: table descriptions, metric definitions, join paths, and sample values. The AI gets rich context about what columns mean and how tables relate, so queries are accurate out of the box.

    One connector for all your data
    Semantic context layer (joins, metrics, descriptions)
    RBAC permissions enforced per-query
    Minutes to set up

    Quick setup

    1

    Sign up at rig.so and connect your data sources (warehouse, CRM, analytics).

    2

    Copy your MCP connector URL from the Rig dashboard. Rig indexes your data, builds a context layer with table descriptions, metric definitions, and join paths, and exposes it all through one MCP endpoint.

    3

    Add the connector in Claude. In Claude Code, add it to your .claude/settings.json under mcpServers. In Cowork, go to Settings → Connectors → Add MCP and paste the URL.

    Direct warehouse connection

    Connect your Snowflake, BigQuery, or Postgres directly as an MCP server. This works, but you'll lack the semantic layer. Column names like STAGE__STATUS__TITLE won't be self-explanatory, so expect more back-and-forth to clarify schema meaning.

    Full SQL access to your warehouse
    No intermediary
    No semantic context (column meanings, join paths)
    Manual setup per warehouse
    More trial-and-error on queries
    Individual tool MCPs

    Connect your CRM, analytics, and other tools as separate MCP servers. Good for smaller companies where the dataset is manageable and you're happy to maintain the connections. Data is typically well-structured but you'll need to stitch it together across sources yourself.

    Works for small, focused datasets
    Direct access to each tool's API
    Good way to get started quickly
    No cross-source joins or context
    Reliability depends on each MCP's quality
    You maintain multiple connections

    Which approach should you choose? If you have a warehouse (Snowflake, BigQuery, Postgres) with CRM and product data flowing into it, Rig gives you the best results because the AI understands how your tables relate to each other. If you're a small team and your data lives entirely in a CRM like HubSpot, a direct CRM MCP can work well to get started. You can always upgrade to a unified approach later.

    3

    Add the skill file

    Skills are markdown files that teach Claude how to perform a specific workflow. Download the board deck builder skill and add it to Claude.

    Download the skill files

    The download includes the board deck builder skill along with its supporting reference files (slide patterns, PptxGenJS reference, etc.).

    Download from Google Drive
    Add to Claude

    In Cowork

    1. Open a new Cowork session

    2. Click the + button or drag-and-drop the skill .md files into the conversation

    3. Claude will recognize them as skill files and use them to guide the deck-building workflow

    In Claude Code

    1. Place the skill files in your project's .claude/skills/ directory

    2. Or use /skill to load them at the start of a session

    3. Claude will pick up the skill files automatically and follow the workflow

    How Claude builds the deck, step by step

    The skill guides Claude through seven phases, from briefing to final delivery. Each phase has a review checkpoint where you approve the data and direction before it moves on. No hallucinated numbers, no guesswork.

    1
    Briefing

    The skill asks about your meeting type, audience, topics, and data sources. It proposes a default slide structure based on your company stage.

    2
    Brand extraction

    Point it at your website, a CSS file, or an existing deck. It pulls your colors, fonts, and visual style into a reusable design token file.

    3
    Data discovery

    It explores your connected data sources, maps tables to deck topics, identifies metrics it can compute, and flags any gaps.

    4
    Deck architecture

    A slide-by-slide plan with layout types, data sources, and visual approach for each slide. You review and adjust before anything gets built.

    5
    Data compilation

    Pulls all the numbers, validates them, and shows you a data summary for approval. No figures go into the deck without your sign-off.

    6
    Build & QA

    Generates the PPTX using PptxGenJS, converts to images, visually inspects every slide, and fixes issues automatically.

    7
    Review & iterate

    Presents the deck with thumbnails. You request changes, it rebuilds. Repeat until it's ready.

    Tips for getting the best results from AI-generated decks

    Provide a previous deck

    If you have a previous board deck or template, share it. The skill can extract the structure, layout patterns, and brand elements as a starting point.

    Connect as many sources as you can

    The more data the skill can access, the richer the deck. CRM for pipeline, warehouse for revenue, analytics for product usage, meeting notes for qualitative insights.

    Review the data summary carefully

    Phase 5 compiles every number that will appear in your deck. This is your chance to catch errors before they end up in front of your board.

    Iterate on specific slides

    After the first build, you can request targeted changes: swap a table for a chart, reorder slides, adjust wording, add or remove sections.

    Build your next board deck in 20 minutes

    Download the skill file, connect your data to Claude, and generate your first deck. Setup takes 10 minutes. The deck takes another 10.