Knowi MCP Server: Build Dashboards from Claude in One Instruction

Introduction

Every major BI platform now has an MCP server. Power BI, Tableau, Looker, Qlik, ThoughtSpot, Domo. Having MCP is table stakes.

The question is what happens after the connection is made.

Most BI platforms expose their existing API operations through MCP. Create a dashboard. Add a widget. Set the chart type. Position it on the grid. Repeat. To build a 3-widget dashboard, your AI tool makes 7 to 10 sequential calls, guessing chart types from a list, guessing layout coordinates, guessing query syntax it has never seen.

Knowi took a different approach.

Table of Contents

Agent-First, Not CRUD-First

Knowi’s MCP server exposes 12 tools instead of 45. The primary tool is knowi_do, which accepts a single natural language instruction and delegates to an internal orchestrator that chains 20+ specialized AI agents.

The agents know the data schema. They know which chart type works for which field combination. They know the query syntax. They handle layout automatically.

Traditional BI MCP (10 calls, AI tool orchestrates):

1. Search for the right dataset

2. Create an empty dashboard

3. Create widget 1 (AI guesses: bar chart?)

4. Create widget 2 (AI guesses: line chart?)

5. Create widget 3 (AI guesses: data table?)

6. Position widget 1 (AI guesses: col 0, row 0?)

7. Position widget 2 (AI guesses: col 6, row 0?)

8. Position widget 3 (AI guesses: col 0, row 4?)

9. Apply theme

10. Take screenshot to check

Knowi MCP (1 call, Knowi orchestrates):

1. knowi_do: “Create a sales dashboard by region

   with a revenue trend and top customers table”

   → Orchestrator detects 4 intents

   → SearchAssets finds the Sales dataset

   → CreateDashboard creates the container

   → CreateWidget runs 3 times:

     – Analyzes schema: region (string), revenue (numeric), date (date)

     – Picks BAR for region x revenue (categorical comparison)

     – Picks LINE for revenue x date (time series)

     – Picks DATA_GRID for detail view

     – Generates Cloud9QL with correct field names

     – Auto-positions widgets on grid

   → DashboardSettings applies layout

   → Returns: dashboard URL, widget details, summary

One call. The orchestrator handled everything internally. Claude does not need to know about chart types, grid coordinates, or query syntax.

The 12 MCP Tools

Knowi’s MCP surface is intentionally small. Fewer tools means better AI performance and fewer hallucinated API calls.

3 Intelligent Tools (Agent-Powered)

9 Deterministic Tools (No AI, Instant, Free)

The intelligent tools cost agent calls (LLM processing). The deterministic tools are unlimited and instant.

Setup: 2 Minutes

Claude Desktop

Add to your Claude Desktop config:

{

  "mcpServers": {

    "knowi": {

      "command": "java",

      "args": ["-jar", "/path/to/knowi-mcp-cli.jar"],

      "env": {

        "KNOWI_URL": "https://www.knowi.com",

        "KNOWI_CLIENT_ID": "<your-client-id>",

        "KNOWI_CLIENT_SECRET": "<your-client-secret>"

      }

    }

  }

}

Restart Claude Desktop. Knowi tools appear in the tool picker.

Claude Code

Add to .claude/settings.json in your project:

{

  "mcpServers": {

    "knowi": {

      "command": "java",

      "args": ["-jar", "/path/to/knowi-mcp-cli.jar"],

      "env": {

        "KNOWI_URL": "https://www.knowi.com",

        "KNOWI_CLIENT_ID": "<your-client-id>",

        "KNOWI_CLIENT_SECRET": "<your-client-secret>"

      }

    }

  }

}

Any MCP-Compatible Tool

The MCP server also exposes a REST API at /api/2.0/mcp/tools/call for tools that support HTTP-based MCP or for custom integrations.

What You Can Do

Ask your data questions

You: “What were our top customers by revenue last month?”

Claude calls knowi_ask and returns:

“Here are your top 10 customers by revenue for February 2026:

 1. Acme Corp — $234,500

 2. GlobalTech — $189,200

 3. …”

Build dashboards from conversation

You: “Create a dashboard for Q4 sales analysis

      with revenue by region, monthly trend,

      and a top sales reps table.”

Claude calls knowi_do. Knowi’s orchestrator:

– Finds the Sales dataset

– Creates the dashboard

– Adds 3 widgets (BAR, LINE, DATA_GRID)

– Auto-positions and applies theme

Returns: dashboard URL + widget summary.

Push data and visualize it

You: “I have some KeyCloak auth event data.

      Push it to Knowi and build me a security dashboard.”

Claude generates realistic auth data, calls knowi_push_data,

then calls knowi_do to create the dashboard.

Claude then calls knowi_screenshot to check how it looks,

and calls knowi_do again to fix layout issues it spots.

Generate embed code for your app

You: “Give me the embed code for that dashboard,

      filtered to tenant ABC.”

Claude calls knowi_get_embed_url with SSO user

and content filters. Returns the embed snippet.

55+ Data Sources, Native Connectors

Knowi connects natively to SQL databases (PostgreSQL, MySQL, Snowflake, BigQuery, Redshift, ClickHouse), NoSQL databases (MongoDB, Elasticsearch, Cassandra, InfluxDB, DynamoDB), REST APIs, S3, Google Sheets, CSV, Excel, and documents.

Native means the agents query each source in its own language. MongoDB gets Mongo queries. Elasticsearch gets Elasticsearch queries. No JDBC flattening, no intermediate warehouse. Cross-source joins happen without moving the data.

Who This Is For

Developers using Claude or any MCP client: Install the MCP server, connect your database, start querying from your AI tool. Free tier available.

SaaS companies with embedded analytics: Embed Knowi’s agentic widget in your app. Your customers ask questions, Knowi returns answers and charts, filtered to their data. White-labeled.

Teams in Slack/Teams: The Knowi bot handles full orchestrator access. /knowi create a sales dashboard by region in Slack triggers the same agent chain as the MCP server.

Enterprise data teams: 20+ agents, private AI, on-premise deployment, SOC 2 Type II, HIPAA-compliant. The same agent system works across all deployment modes.

Get Started

Knowi Enterprise: MCP server included in Enterprise and On-Premise plans. Request a Demo.

Frequently Asked Questions