AI Tools Quickstart
Massive delivers a wide range of high-quality U.S. market data that is well suited for AI agents and LLM powered applications. Our REST API, real-time WebSocket streams, Flat Files, official MCP server, and LLM optimized documentation (llms.txt) give you multiple ways to connect financial data to intelligent systems.
This guide explains the different integration approaches, when to use each one, and how to achieve the best results whether you are building with Claude, OpenAI, custom agents, or other frameworks.
For raw data access patterns, see the REST Quickstart, WebSocket Quickstart, and Flat Files Quickstart. This page focuses on the higher-level patterns for turning that data into intelligent, agentic workflows.
MCP Server
The Massive MCP Server is the fastest and most effective way for AI agents to work with our data. Instead of exposing dozens of individual endpoints as tools (which consumes excessive context), it provides a small set of powerful, composable tools that let agents:
- Discover relevant endpoints using natural language search
- Retrieve accurate parameter documentation on demand
- Execute any REST API call
The server dynamically indexes our documentation from llms.txt at startup, so it remains automatically up to date as we add new endpoints and features.
This design delivers dramatically lower token usage compared to traditional tool per endpoint approaches, along with better tool selection accuracy and strong support for complex, multi-step financial analysis workflows. It works especially well with Claude Code, Claude Desktop, Cursor, and other MCP-compatible clients.
Massive offers the MCP server two ways. Most users should start with the Remote MCP Server: it is hosted by Massive, so once you configure your client and authenticate over OAuth there is nothing else to install or maintain. The self-hosted option runs in your own environment and is worth considering mainly when you need to pull large datasets locally.
Remote MCP Server (Recommended)
The Remote MCP Server is hosted at https://mcp.massive.com/ and authenticates over OAuth. There is nothing to install. You point your client at the URL, sign in once, and start querying. This is the easiest path and the right choice for nearly everyone.
The hosted environment currently provides 64 MB of working storage per session and supports up to 10 concurrent workspaces (roughly 10 simultaneous chats). That is ample for typical agentic analysis, but if you routinely download very large datasets the self-hosted option may be a better fit.
Access through the MCP server mirrors your Massive entitlements. You can query any asset class your plan includes. If you have Stocks access, stocks data is available through MCP. Requesting data you are not entitled to (for example, Options without an Options plan) returns an access-denied error.
Each client connects a little differently. Pick your tool for setup, verification, and troubleshooting tailored to it:
Self-Hosted MCP Server
If you need to work with large local datasets or prefer to run everything in your own environment, you can self-host the server. On top of the core tools, it adds a local data workspace that can store results as in-memory DataFrames, run SQL queries against them, and apply built-in financial functions (Greeks, returns, technical indicators, and more), which is useful for heavier local analysis. Install it via uv or pip from the official repository: github.com/massive-com/mcp_massive.
LLM Friendly Documentation
Massive's documentation is built to be AI-friendly from the ground up. Every page can be accessed as clean, structured Markdown by appending .md to the URL. This is the same content served by the "Open as Markdown" button in the top right corner of endpoint pages.
We also provide llms.txt and llms-full.txt files at any level of the documentation navigation tree. You can append these files to most paths, from the root down to section or category levels. This gives you flexible control: move higher in the tree for a broader view, or lower for a more tightly scoped index.
Examples of how to scope these files:
| Level | llms.txt (Compact Index) | llms-full.txt (Expanded Details) |
|---|---|---|
| Root | /docs/llms.txt | /docs/llms-full.txt |
| REST Section | /docs/rest/llms.txt | /docs/rest/llms-full.txt |
| Stocks Section | /docs/rest/stocks/llms.txt | /docs/rest/stocks/llms-full.txt |
| Specific Category | /docs/rest/stocks/tickers/llms.txt | /docs/rest/stocks/tickers/llms-full.txt |
Choosing the right file:
- llms.txt: Best for limited context windows and general discovery (lightweight index)
- llms-full.txt: Best when the agent needs detailed endpoint descriptions, parameters, and response schemas
This tree based system enables AI agents to pull exactly the right amount of documentation context they need, whether a high-level overview, or focused details on a particular area of the API.
Direct REST API Integration
You can also connect AI systems directly to our REST API using function calling and tool use in frameworks like OpenAI and Anthropic. Many developers now start by describing their desired workflow in natural language and letting AI generate the initial code. This approach works especially well for rapid prototyping and exploration.
We strongly recommend using our official client libraries as the foundation when building production code:
- Python: github.com/massive-com/client-python
- JavaScript: github.com/massive-com/client-js
- Go: github.com/massive-com/client-go
- Kotlin / JVM: github.com/massive-com/client-jvm
These libraries handle authentication, retries, pagination, error handling, and response parsing, giving you a solid, production-ready foundation, even when AI helps generate surrounding application code.
Our documentation pages include ready to use code examples for all endpoints. You can combine these examples with our llms.txt or .md files and feed them into your AI coding assistant to generate accurate, well structured integrations quickly. This combination of, strong client libraries + high-quality documentation + AI assistance, gives you the best balance of speed and reliability.
Skills & Instructions
You can significantly improve AI performance by leveraging Skills (or equivalent project instructions and prompt libraries) in your AI tools. These act as a powerful shortcut, they teach the AI how your team specifically works with market data, rather than leaving it to guess.
A Skill encodes your internal conventions and best practices so the AI produces more consistent, accurate, and compliant outputs.
Common examples include:
- Rules for using adjusted vs. unadjusted prices
- Data quality checks and internal compliance guidelines
- Handling of corporate actions, date ranges, and market hours
- Preferred calculation methods for yields, returns, or volatility
You can implement this through Claude Skills (SKILL.md files), persistent project instructions, or reusable prompt libraries. This layer sits on top of your data access method (MCP or REST) and is one of the highest leverage ways to get reliable results from AI.
Choosing the Right Approach
Choosing the right integration method depends on your goals, the complexity of your workflows, and how much control you need. The table below summarizes the main options and when each one tends to work best.
| Approach | Best For | Context Efficiency | Setup Effort | Recommended When |
|---|---|---|---|---|
| Massive MCP Server | Most agentic workflows & complex analysis | Excellent | Low | You want agents to discover and use data intelligently |
| llms.txt + Markdown | Documentation-aware agents and discovery | Very Good | None | Agents need accurate, up to date API reference |
| Direct REST + Function Calling | Custom orchestration or full control | Good | Medium | Building your own agent framework or needing fine-grained control |
| Skills / Instructions | Consistent team processes & domain knowledge | Excellent | Low | You have repeatable financial workflows or compliance rules |
Quick decision guide:
- Building or using an agent that needs to take actions on live data? Start with the MCP Server.
- Need accurate, current endpoint documentation in context? Use llms.txt.
- Have specific internal processes or rules the AI must follow? Add Skills or project instructions.
- Want maximum control or custom orchestration? Use Direct REST with function calling + client libraries.
Next Steps
The fastest way to get value is usually to start with the MCP Server and llms.txt, then layer in Skills as your needs grow.
- Pick the client that matches your tool under MCP Server and follow its setup guide.
- Explore our llms.txt files so your agents can discover documentation automatically.
- Review the REST Quickstart if you need to understand authentication and basic request patterns.
With these tools in place, you can move beyond manual data copying and start building reliable, production grade financial intelligence with AI.