MCP server

A Model Context Protocol (MCP) server is a lightweight service that exposes data, tools, and context to AI agents and large language models (LLMs) in a standardized way. Originally introduced by Anthropic in late 2024, MCP defines an open protocol that lets AI systems connect to external data sources and tools without requiring custom, one-off integrations for every combination of model and data source. Think of it as a universal adapter: instead of building a bespoke connector every time you want an AI agent to talk to a new system, you build one MCP server and any MCP-compatible client can use it.

Here is a concrete way to picture it. Say your analytics team has an AI assistant that needs to pull live sales data from your data warehouse, check a customer record in your CRM, and run a calculation against a financial model, all in a single workflow. Without MCP, each of those connections requires its own integration logic. With an MCP server sitting in front of each system, the AI agent sends a standardized request, the MCP server handles the translation and retrieval, and the agent gets back structured context it can actually use. A team that previously spent weeks wiring up three separate integrations can now do it in a fraction of the time.

The gap between what AI agents are theoretically capable of and what they can actually do in production often comes down to data access. An LLM is only as useful as the context you give it, and most enterprise data lives in systems that AI tools cannot natively reach. MCP servers close that gap by giving AI agents a reliable, governed path to the data and tools they need to complete real tasks. For analytics teams specifically, this means AI assistants can move beyond answering questions about static datasets and start interacting with live, operational data. That shift, from AI as a query interface to AI as an active participant in your data workflows, is what makes MCP worth paying attention to right now.

1. Define the resources and tools you want to expose, like database tables, API endpoints, or analytical functions, inside the MCP server configuration.

2. Register the MCP server with an MCP-compatible AI client or agent framework so the client knows where to route requests.

3. Receive a structured request from the AI agent specifying what context or action it needs, formatted according to the MCP specification.

4. Authenticate and authorize the request against your existing access controls before returning any data.

5. Retrieve the requested data or execute the specified tool and return a structured response the AI agent can parse and act on.

6. Log the interaction for auditing and governance purposes, maintaining a record of what data the AI accessed and when.

• A global retail company connects its data warehouse to an AI agent through an MCP server, giving the agent read access to daily sales figures, inventory levels, and promotional performance data. When a merchandising analyst asks the agent why a product category underperformed last week, the agent queries the warehouse in real time through the MCP server, pulls the relevant rows, and surfaces a clear answer with supporting numbers, cutting the analyst's investigation time from two hours to under five minutes.

• A financial services firm uses an MCP server to give its AI assistant access to a portfolio management system and a real-time market data feed. Compliance analysts can ask the assistant to flag any positions that breach concentration limits as of the current trading day. The MCP server handles authentication, retrieves the live data, and returns a structured list of flagged positions, a task that previously required a manual pull from two separate systems and a spreadsheet merge.

• A SaaS company's customer success team connects their CRM and product usage database to an AI agent via two separate MCP servers. When a customer success manager asks the agent to identify accounts showing early signs of churn, the agent queries both systems through their respective MCP servers, correlates login frequency drops with open support tickets, and returns a ranked list of at-risk accounts. The team reports a 30% improvement in the speed of their weekly churn review process after deploying this setup.

1. Standardized integration: Instead of writing custom glue code for every AI-to-data-source connection, you build one MCP server per system and any compatible AI client can use it. A data engineering team that manages ten internal tools no longer needs to maintain ten separate integration scripts for each AI assistant the organization adopts.

2. Governed data access: MCP servers sit between the AI agent and your data, which means you can enforce authentication, row-level security, and audit logging at the server layer. This gives your security and governance teams a single, inspectable control point rather than hoping each AI tool handles access correctly on its own.

3. Faster AI agent development: Developers building AI agents spend less time on plumbing and more time on the actual logic of the agent. Because the MCP specification is open and consistent, a developer who has built one MCP client integration can connect to any MCP server without learning a new API contract each time.

4. Live context for AI workflows: MCP servers retrieve data at request time rather than relying on pre-loaded, potentially stale context. For analytics use cases, this means an AI agent answering a question about revenue trends is working with today's numbers, not last week's snapshot.

5. Composability across systems: A single AI agent can call multiple MCP servers in one workflow, combining context from a data warehouse, a CRM, and a business intelligence platform in a single response. This composability is what makes agentic analytics workflows practical rather than theoretical.

ThoughtSpot's MCP server makes it possible for AI agents and LLM-powered applications to query your ThoughtSpot data models, Liveboards, and Answers directly through the MCP protocol. When you connect an AI assistant to ThoughtSpot via MCP, it can retrieve trusted, governed analytics context, the same metrics and definitions your analysts rely on, and bring that context into any MCP-compatible workflow. Spotter, ThoughtSpot's AI analyst, already operates on the principle that AI should work with your verified data layer rather than around it, and MCP extends that principle to the broader ecosystem of AI agents your organization is building. For teams using ThoughtSpot Embedded, MCP opens up new ways to surface analytics context inside the AI-powered products you are shipping to your own customers.

1. Search-Based Analytics

2. Conversational Analytics

3. Self-Service Analytics

4. Large Language Models (LLMs)

5. Prompt Engineering

6. Semantic Layer

7. AI-Powered Analytics

An MCP server is a standardized service that gives AI agents governed, real-time access to your data and tools, replacing fragile one-off integrations with a consistent protocol that any MCP-compatible client can use.