Knowi’s MCP server lets external AI tools (Claude, GPT, GitHub Copilot, Cursor) connect directly to your analytics platform and take action on it. Using the Model Context Protocol standard, any MCP-compatible AI tool can query your databases, create dashboards, search assets, and generate reports through natural language, without logging into Knowi’s UI. MCP is the protocol layer that turns your BI platform from a standalone application into an API that any AI assistant can operate.
TL;DR
- MCP (Model Context Protocol) is an open standard by Anthropic that lets AI tools connect to external systems through a standardized interface.
- Knowi is the first BI platform with a production MCP server. Claude and any MCP-compatible tool can query your data and build dashboards without opening Knowi’s UI.
- Three core operations:
knowi_ask(natural language queries),knowi_do(create dashboards, widgets, exports), andknowi_search(find existing assets). - One MCP invocation = one agent call for billing, even if five internal agents chain together behind the scenes.
- Works with Private AI deployment: the MCP server runs inside your environment, so no data leaves your infrastructure.
- Latency: 5-15 seconds per orchestrator call, with streaming support for real-time responses in terminals and IDEs.
Table of Contents
What Is MCP?
Model Context Protocol is an open standard created by Anthropic that defines how AI models connect to external tools and data sources. Think of it as USB-C for AI: a single, standardized interface that any AI tool can use to plug into any compatible system.
Before MCP, connecting an AI tool to your analytics platform meant building custom integrations: API wrappers, authentication flows, response parsing. Every AI tool needed its own connector. MCP replaces all of that with one protocol that every MCP-compatible tool speaks natively.
How MCP Differs from APIs
A REST API exposes endpoints that a developer calls with specific parameters. MCP exposes capabilities that an AI model discovers and uses autonomously. The AI tool reads the MCP server’s capability description, understands what it can do, and decides when and how to use it based on the user’s natural language request.
You don’t write API calls. You say “show me revenue by region last quarter” and the AI tool figures out which MCP operations to invoke.
How Knowi’s MCP Server Works
Three Core Operations
knowi_ask: Natural language queries against any connected data source. “What were the top 10 products by revenue last month?” The Query Agent identifies the right database, writes the query, and returns results. Works across SQL, NoSQL, and API sources without the user knowing which database holds the data.
knowi_do: Action commands that create or modify analytics assets. “Create a dashboard showing customer churn by segment.” The orchestrator invokes the Dashboard Agent, Widget Agent, and any other agents needed to fulfill the request. Also handles exports, share links, and report scheduling.
knowi_search: Semantic search across all Knowi assets. “Find dashboards related to Q4 sales.” Uses vector search (Milvus) to find relevant dashboards, widgets, datasets, and queries by meaning, not just keyword matching.
The Orchestration Behind Each Call
When an MCP request arrives, it goes through the same AI Agent Orchestrator that powers in-platform interactions. The orchestrator:
- Receives the natural language request from the external AI tool
- Uses LLM intent detection to determine which agents to invoke
- Executes up to five specialized agents in priority order
- Passes context between agents (results from agent 1 are available to agent 2)
- Aggregates results and returns them to the AI tool
A single knowi_do call might internally chain Query Agent, Dashboard Agent, Widget Agent, and Recommendation Agent. From a billing perspective, this counts as one agent call.
Use Cases: MCP in Practice
Developer in an IDE
A developer using Claude connects the Knowi MCP server. Without leaving their editor, they ask “what’s the error rate trend for the payments service this week?” The Query Agent pulls data from their Elasticsearch cluster and returns a summary directly in the IDE chat panel.
Analyst in Claude Desktop
An analyst opens Claude Desktop and asks “create a dashboard comparing Q4 and Q1 revenue by product line.” Claude invokes knowi_do, the orchestrator creates the dashboard with appropriate charts, and Claude returns the dashboard URL. The analyst clicks through to a live, interactive dashboard.
Executive on Slack
Through the Slack integration (which uses the same agent orchestration), a VP asks “how did the healthcare segment perform last month?” The system queries across their CRM, billing database, and usage analytics, blends the results, and posts a summary with a link to the full dashboard.
SaaS Product with Embedded MCP
A SaaS company embeds Knowi’s agentic widget in their product. Their customers’ AI tools (Claude or any MCP client) can connect via MCP to query only their own tenant’s data, enforced by row-level security. Each customer gets an AI-powered analytics experience without the SaaS company building any AI infrastructure.
MCP for Analytics: Platform Comparison
| Capability | Tableau | Power BI | ThoughtSpot | Knowi |
| MCP server | No | No | No | Yes (production, first in BI) |
| External AI tool access | REST API only (developer-built) | Copilot (Microsoft ecosystem only) | API only | Any MCP-compatible tool: Claude or any MCP client |
| Natural language to dashboard | Ask Data (limited to current view) | Copilot (current report only) | SearchIQ (requires semantic model) | Creates dashboards from any AI tool via MCP, no model required |
| Multi-source queries | Requires extract or warehouse | Requires data model in Power BI | Requires semantic layer (weeks to build) | Queries SQL, NoSQL, APIs natively, cross-source joins without ETL |
| Agent orchestration | None | Single Copilot model | Single NLQ model | Up to 5 specialized agents per request, coordinated by orchestrator |
| Private deployment | Tableau Server (on-prem) | Power BI Report Server | Cloud only | Full on-prem: MCP server, agents, LLM all inside your environment |
| Embedded + MCP | Not available | Not available | Not available | White-label embedded with per-tenant MCP access |
Technical Setup
Connecting Claude Desktop to Knowi
Add the Knowi MCP server to your Claude Desktop configuration file. Provide your Knowi API credentials and specify which data sources the AI tool can access. Claude automatically discovers available operations (ask, do, search) and uses them when your questions require analytics data.
Connecting from an IDE
Other MCP-compatible tools support the same configuration. Point the MCP server URL to your Knowi instance (cloud or on-prem), authenticate, and start querying data from your AI tool’s chat panel. See the Knowi AI tools documentation for setup steps.
Security and Access Control
MCP requests inherit Knowi’s existing permission model. Row-level security, role-based access, and data source restrictions all apply. An MCP user can only query data and dashboards they have permission to access. For Private AI deployments, the MCP server runs entirely on-prem, with no external network calls.
Latency and Billing
MCP requests go through the agent orchestrator, which takes 5-15 seconds per call depending on complexity. Streaming is supported for real-time token delivery in terminals and IDEs.
Billing is straightforward: one invocation of knowi_ask, knowi_do, or knowi_search equals one agent call. Even if the orchestrator chains five internal agents to fulfill the request, it counts as a single call.
Frequently Asked Questions
What is MCP and why does it matter for analytics?
MCP (Model Context Protocol) is an open standard by Anthropic that lets AI tools connect to external systems through a single, standardized interface. For analytics, it means any AI assistant (Claude or any MCP client) can query your databases, build dashboards, and manage reports without custom API integration. Knowi is the first BI platform with a production MCP server.
Which AI tools can connect to Knowi via MCP?
Any MCP-compatible tool: Claude Desktop, Claude Code, and any future tools that adopt the MCP standard. The protocol is open, so adoption is growing rapidly across the AI ecosystem.
Do I need to write code to use Knowi’s MCP server?
No. You add the Knowi MCP server to your AI tool’s configuration (a JSON file), authenticate, and start asking questions in natural language. The AI tool discovers Knowi’s capabilities automatically and uses them when your requests require analytics data.
Can MCP access NoSQL databases like MongoDB and Elasticsearch?
Yes. Knowi connects natively to 55+ data sources including MongoDB, Elasticsearch, Cassandra, DynamoDB, PostgreSQL, MySQL, Snowflake, and REST APIs. MCP requests go through the same Query Agent that handles in-platform queries, so all native connectivity is available.
Is data sent to external servers when using MCP?
With Knowi’s Private AI deployment, no. The MCP server, agent orchestrator, and LLM all run inside your environment. Queries, results, and dashboard data never leave your infrastructure. For cloud deployments, data is processed within Knowi’s SOC 2 Type II certified environment.
Can my SaaS customers use MCP to access their own analytics?
Yes. Knowi supports per-tenant MCP access through its embedded analytics platform. Each customer connects their AI tool to your Knowi-powered analytics layer, with row-level security ensuring they only see their own data. This is available on Knowi Enterprise plans.
