TwelveLabs' approach

TwelveLabs is a video understanding platform that enables developers and enterprises to search, analyze, and generate insights from video content using AI. Their platform combines proprietary models (Marengo for multimodal embeddings and Pegasus for video-language understanding) and powers capabilities that include semantic video search, automatic summarization, content analysis, and embedding generation.

Rather than requiring users to build bespoke video processing pipelines, TwelveLabs packages these capabilities as API-accessible services. Developers can index videos, search through footage using natural language, generate structured summaries, and extract metadata. This can all be done without having to handle the complexity of video processing under the hood.

As AI assistants became the primary interface for developer workflows, TwelveLabs saw an opportunity to make video intelligence a first-class capability inside any MCP-compatible environment, from Claude and Cursor to custom AI agents.

The challenge

Video remains one of the most underutilized data types in AI workflows. While LLMs excel at processing text, they have historically been blind to video content. Developers who wanted to integrate video understanding into their applications faced several obstacles.

Custom API integration was complex and time-consuming. Each AI environment required its own glue code, authentication handling, and response parsing. There was no standardized way for an LLM agent to discover and invoke video tools at runtime.

Beyond integration complexity, video workflows are inherently multi-step. A meaningful task, such as finding specific moments in a video library, analyzing them, and producing a summary requires chaining multiple operations together. Without a protocol-level standard for tool discovery and invocation, orchestrating these pipelines demanded significant engineering effort.

TwelveLabs needed a distribution strategy that would make their video intelligence accessible inside any AI assistant, without requiring developers to write custom plugins or integration code for each platform.

The solution

TwelveLabs built an MCP server that exposes their full video understanding platform (indexing, semantic search, analysis, and embeddings) as standardized MCP tools, resources, and prompts. Any MCP-compatible client can discover and use these capabilities out of the box.

As shown in the image below, Claude can search a library of fashion videos and return exact timestamps for every relevant segment across multiple videos, no custom code, just a natural language prompt.

The server enables AI assistants to perform powerful video tasks on demand. Users can ask an agent to find specific moments in hours of footage or summarize lengthy meeting recordings in seconds.

Crucially, the TwelveLabs MCP server is designed to work as a composable building block alongside other MCP servers. TwelveLabs has demonstrated advanced workflows combining their server with third-party MCP integrations, such as:

• TwelveLabs + Jotform MCP: Analyze video content to auto-generate quizzes and assessments, then publish them as shareable live forms.

• TwelveLabs + Exa MCP: Extract topics and claims from video, then search the web for citations and source links, enabling automated fact-checking.

• TwelveLabs + Sora 2 MCP: Generate video assets with Sora 2, then use TwelveLabs to verify brand consistency and guide editing decisions for ads and social content.

This composability is a direct benefit of the MCP standard: because each server follows the same protocol, developers can stack them together without writing custom glue code between services.

Alpic’s role

TwelveLabs partnered with Alpic to build, host and distribute their MCP server in production. After assisting in the design and creation of the MCP server, Alpic provided the infrastructure that makes the TwelveLabs MCP server accessible, reliable, and easy to install.

For TwelveLabs, Alpic delivers: