Pyth Pro for AI Agents: Institutional Market Data for Autonomous Finance

The agentic AI market in financial services is projected to grow from $5.2 billion in 2024 to nearly $197 billion by 2034. 44% of finance teams plan to deploy agentic AI in 2026, up over 600% from 2025. Bloomberg has embedded LLMs directly into its terminal. The top 50 banks announced more than 160 agentic AI use cases in a single year.

AI agents will consume more market data than humans ever have. A single agent workflow can query prices across dozens of assets in seconds, repeatedly, 24/7. Multiply that by every institution deploying agentic workflows, and the demand for programmatic, cross-asset market data dwarfs anything the terminal era produced.

Today, Pyth launches Pyth Pro for AI Agents to meet that demand: an agent-native market data service that gives any AI agent access to 3,000+ institutional-grade price feeds across crypto, equities, FX, metals, and commodities. Built on the same data infrastructure that secured $2.7 trillion in transaction volume across 700+ onchain and offchain applications.

This is not an AI app. It is infrastructure for the era of autonomous finance.

Why the existing data stack cannot serve agents

When a human analyst pulls a price from a terminal, they apply judgment. They cross-check, they notice when something looks off, they read the context around the number. Agents operating at scale cannot do any of that. A bad input propagates through the workflow and produces a bad output. There is no friction in the loop to catch it.

This makes data quality even more consequential for agents than for humans. Today, most agent workflows pull financial data from web search or public APIs with opaque methodologies. These sources were not designed for programmatic use at scale. They introduce latency, inconsistency, and coverage gaps that would be immediately disqualifying in any institutional context.

The second problem is fragmentation. An AI agent assigned to monitor cross-asset conditions needs prices across equities, FX, crypto, and commodities. In the traditional stack, that means four different vendor integrations, four different authentication flows, and four different data formats to normalize. For an agent workflow, that is an architectural constraint that limits what the agent can do.

The third problem is structural. The largest market data providers built their infrastructure for human users sitting at proprietary terminals. Their licensing models reflect that: restrictive redistribution terms, per-seat pricing, and usage constraints that prohibit programmatic consumption by third-party agents. The delivery model and legal framework were not built for autonomous consumption at scale.

Why Pyth is uniquely positioned

Pyth's architecture was designed, from the ground up, for permissionless programmatic consumption. This is not a feature that was added later. It is how the network was built.

First-party data from the firms that set prices. Pyth's prices come from institutions that are active participants in price discovery: trading firms, exchanges, market makers, and banks. The data does not pass through intermediaries who repackage, delay, or mark it up. When the consumer is an autonomous system making real-time decisions, the pedigree of the input determines the quality of every output downstream.

Redistribution-friendly licensing. Pyth Pro's licensing permits programmatic access and redistribution to downstream systems, including AI agents. This is the structural difference that matters most. Legacy providers restrict how their data can be consumed and by whom. Pyth Pro AI allows any agent to access the data, build on it, and feed it into autonomous workflows without triggering licensing violations. For institutions deploying agentic systems in production, it is a prerequisite.

Every asset class, one integration. 3,000+ price feeds spanning crypto, equities, FX, metals, and commodities, accessible through a single connection. Assembling equivalent coverage through traditional vendors requires multiple contracts, multiple authentication flows, and multiple data formats. For an agent that needs to reason across asset classes in real time, Pyth eliminates the integration burden entirely.

Open protocol, open source. Pyth Pro AI is open source and available wherever agents operate. It is built on MCP (Model Context Protocol), the emerging standard for connecting AI agents to external tools and data, and listed across every major MCP registry: Claude, Cursor, Glama, and the Solana ecosystem. There is no lock-in to a specific AI platform, and no requirement to operate inside a proprietary terminal or interface. The data goes where the agents are.

No other market data provider combines all four: first-party institutional data, redistribution-friendly licensing, unified cross-asset coverage, and open-protocol delivery. Other oracle networks have open delivery but source from the same third-party APIs that agents can already access directly. Pyth is the only provider where the architecture, the data, and the licensing are all built for a world where agents are the primary consumers.

How Pyth Pro for AI Agents works

Pyth Pro AI is built on MCP (Model Context Protocol), the emerging standard for connecting AI agents to external tools and data. MCP is supported natively by Claude, Cursor, and a growing list of AI platforms. Any agent running on an MCP-compatible client can call Pyth Pro AI tools directly. No custom integrations, no bespoke APIs, no maintenance overhead.

Pyth Pro AI exposes four tools:

get_symbols Search and discover available price feeds by name, asset type, or keyword. 3,000+ feeds spanning crypto, equities, FX, metals, and commodities.

get_latest_price Real-time prices with confidence intervals, updated at millisecond frequency. Requires a Pyth Pro access token.

get_historical_price Prices at specific past timestamps. Historical data available from April 2025 onward.

get_candlestick_data OHLC data at configurable resolutions (1-minute through monthly) for charting, technical analysis, and backtesting.

Three of the four tools work without authentication and are available for free. Real-time pricing via get_latest_price requires a Pyth Pro access token, available at pyth.network/pricing.

How to get started

Pyth Pro AI is available now. Install it directly through Claude or Cursor by searching for Pyth in the MCP marketplace, or clone the repository and configure it manually.

GitHub: github.com/pyth-network/pyth-mcp

Claude MCP registry: Search "Pyth" in the Claude MCP settings

Cursor plugin: Available in the Cursor plugin directory

npm: @pyth-network/mcp-server

For institutions evaluating Pyth Pro as a data layer for production agentic workflows, reach out at pyth.network/contact.

The thesis holds

Pyth started as infrastructure for DeFi, securing over $2.7 trillion in derivatives volume. Pyth Pro extended that same institutional-grade data to traditional finance through a subscription service. Pyth Pro AI extends it to autonomous finance: making that data accessible to the agents and workflows that are becoming the next interface layer in finance.

When AI agents become the largest consumers of market data, the providers that win will be the ones whose architecture was built for programmatic, permissionless consumption from day one. Not the ones trying to retrofit terminal-era licensing for an agentic world.

Pyth was built for this.