AI Agents is a consumer surface, not a separate product. Everything an
agent can do is also available over REST, GraphQL, WebSocket, Kafka, the
SDKs and the CLI. The pages in this section describe the shapes that make
the catalog usable from LLMs.
Two main surfaces
MCP Server
A hosted (and self-hostable) Model Context Protocol server that exposes
ChainStream’s catalog as typed tools. Works out-of-the-box with Claude,
Cursor, ChatGPT, Windsurf and any MCP-compatible client.
Agent Skills
Installable SKILL.md packs that teach an agent how to use
ChainStream — three skills cover standard data (
chainstream-data),
custom GraphQL analytics (chainstream-graphql) and DeFi execution
(chainstream-defi). Ships ready for Cursor, Claude Code and Codex.Why this matters for agents
Traditional LLM access to blockchains fails in three ways: the model has a knowledge cutoff, it has no authenticated data source, and it cannot pay for reads without a pre-registered account. ChainStream solves each:Live data, typed for tool use
Live data, typed for tool use
Every data product is reachable as a structured tool call with a JSON
schema. The agent gets canonical IDs, not screen-scraped text — so it can
chain calls without re-parsing on every step.
Authenticated access that fits agent loops
Authenticated access that fits agent loops
API keys work, OAuth works, and for autonomous agents there is
x402 / MPP — a
per-call HTTP payment scheme that means an agent can pay in USDC and
call immediately, with no prior account setup.
Optional write paths with user-level guardrails
Optional write paths with user-level guardrails
DEX quotes, routing and swap submission are all available. Writes still
require a signed transaction from a user-controlled wallet — agents can
prepare transactions but not silently send from someone else’s funds.
Which surface should I pick?
| Situation | Surface |
|---|---|
| You are using Claude / Cursor / ChatGPT and want ChainStream tools | MCP Server |
| You want to install a ready-made agent skill for token research | chainstream-data skill |
| You need cross-cube JOINs, custom aggregations or time-series | chainstream-graphql skill |
| You want the agent to execute swaps / create tokens | chainstream-defi skill |
| You are writing your own agent in Python / TS and want direct access | SDKs |
| You want pay-per-call with USDC and no signup | x402 Payments |
| You want to pay with Tempo / Stripe / stablecoins | MPP Payments |
| You want shell-native tools the agent can orchestrate | CLI |
Quickstart for agents
Point your agent at it
Full configuration snippets for Claude Desktop, Cursor and ChatGPT are in
MCP Server Setup.
Next
MCP Server introduction
How the MCP Server is organized and what tools it exposes.
Agent Skills introduction
What Skills are, how they are installed and how to author new ones.
x402 Payments
HTTP-native micropayments for agent workloads.
AI Agent tutorial
Build an agent end-to-end using ChainStream.