# Use BlazePhoenix from AI agents: Claude Code, Gemini CLI and any MCP client

> Canonical: https://blazephoenix.xyz/learn/use-blazephoenix-from-ai-agents/
> License: CC BY-NC-ND 4.0 (attribution + link) · © 2026 BlazePhoenix
> Updated: 2026-10-05

How to give an AI agent BlazePhoenix quotes with one command: as a Claude Code plugin, a Gemini CLI extension or an MCP server, hosted with no key or RPC call, or local against your own node. Which other clients can use it, and why the agent never holds keys or sends transactions.

BlazePhoenix is available to AI agents as an MCP server, a Claude Code plugin and a Gemini CLI extension. Each install is 1 command. The hosted server needs no key and makes no RPC call; the local server reads the chain through your own node and adds one-step quotes and swaps.

## How do you add it to Claude Code?

Two routes: the plugin, which bundles the hosted server and a skill, or the MCP server directly.

```
# plugin (hosted MCP + the blazephoenix-defi skill)
claude plugin marketplace add blazephoenixxyz-crypto/blazephoenix-mcp
claude plugin install blazephoenix@blazephoenix

# or the hosted MCP server alone
claude mcp add --transport http blazephoenix https://blazephoenix.xyz/mcp
```

## How do you add it to Gemini CLI?

The extension is published from the same public repository, which carries a gemini-extension.json and a GEMINI.md telling the model how to use the tools.

```
gemini extensions install https://github.com/blazephoenixxyz-crypto/blazephoenix-mcp
```

## When should you run the local server instead?

The hosted server prepares and decodes, and leaves the chain read to the agent. When you want the server to do the read — quotes, swaps and solvency checks in one tool call — run it locally against your own RPC. Your node, your rate limits, no third party in the path.

```
claude mcp add blazephoenix -e BLAZEPHOENIX_RPC_BASE=<your Base node URL> -- npx -y @blazephoenix/mcp
```

## Which other clients can use it?

Any client that speaks the Model Context Protocol over streamable HTTP can point at https://blazephoenix.xyz/mcp. The server is listed in the official MCP registry as xyz.blazephoenix/mcp, which is where registry-aware clients and directories find it. Agents that read the web rather than call tools can start from /llms.txt, the skill file at /skills/blazephoenix/SKILL.md, and the OpenAPI document at /api/openapi.json.

## What will the agent never do?

BlazePhoenix's agent tools are read-only by design.

Hold keys or send transactions for the user. Every tool is read-only; quotes come back with unsigned calldata that the user signs in their own wallet, and the BlazePhoenix Router enforces the minimum output on-chain whatever the agent does. That separation is what makes it safe to give an agent the tools at all.

**Verify it yourself:** curl -s -X POST https://blazephoenix.xyz/mcp -H 'content-type: application/json' -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

Related: https://blazephoenix.xyz/learn/zero-rpc-mcp-server/ · https://blazephoenix.xyz/learn/quote-api-in-five-minutes/ · https://blazephoenix.xyz/learn/api-sdk-mcp/ · https://blazephoenix.xyz/learn/clean-protocol-for-agents/ · https://blazephoenix.xyz/learn/signing-vs-sending/
