Advanced

MCP for AI Agents

Connect BoltTx to any MCP-compatible AI client. Your agent reads plain English — BoltTx delivers the transaction.

Why MCP?

MCP (Model Context Protocol) is an open protocol for connecting AI agents to external tools. BoltTx was the first Solana transaction relay to ship a native MCP server — meaning your agent can inspect your account, look up tip minimums, and submit signed transactions without writing a single line of glue code.

Your private keys never leave your machine

The MCP server only accepts transactions that are ALREADY signed. Signing happens locally using whatever workflow you already use — Phantom, Solflare, a local keypair file, hardware wallet, or your own signing service. The MCP server's job is delivery, not custody.

Install

No package installation needed

You do NOT need to run `npm install @bolttx/mcp-server`. The config snippet below uses `npx -y` which downloads the server to your npm cache on first use (~2-3 seconds) and runs it from cache on every subsequent launch. Updates happen automatically — next time you restart your agent it pulls the latest version.

Prerequisites

Node.js 18 or newer installed (check with `node --version`)
Any MCP-compatible AI agent

Add one entry to your MCP client config — that's it. Your MCP-compatible AI agent automatically spawns the server via `npx` the first time you use a BoltTx tool.

config.json

{
  "mcpServers": {
    "bolttx": {
      "command": "npx",
      "args": ["-y", "@bolttx/mcp-server"],
      "env": {
        "BOLTTX_API_KEY": "btx_live_your_api_key_here"
      }
    }
  }
}

Advanced: install globally instead of using npx

If you prefer a pinned installation (skip the npx cache lookup on every agent restart, or work offline), install the package globally:

shell

npm install -g @bolttx/mcp-server

Then change your MCP config to invoke the installed binary directly instead of through `npx`:

config.json (global-install variant)

{
  "mcpServers": {
    "bolttx": {
      "command": "bolttx-mcp-server",
      "env": {
        "BOLTTX_API_KEY": "btx_live_your_api_key_here"
      }
    }
  }
}

Available tools

These are the tool names your AI agent sees. The agent decides which to call based on your plain-English request.

Parameter	Description
bolttx_get_my_account	Inspect plan, TPS budget, cumulative tip, and upgrade progress for the current API key.
bolttx_list_tip_addresses	Returns the 9 BoLt1-9 vanity tip addresses plus a recommended one (round-robin load distribution).
bolttx_get_tip_amount	Looks up minimum tip (lamports + SOL) for a given plan name.
bolttx_send_transaction	Submits a single fully-signed base64 transaction. Server NEVER signs — agent must sign locally first.
bolttx_send_batch	Submits 1..100 pre-signed transactions concurrently. Per-tx status reported independently.
bolttx_get_status	Looks up BoltTx delivery telemetry for a signature your account submitted. Returns 404 for foreign signatures.

Resources

Read-only context the agent can pull when it needs background. Useful for "what plan am I on?" or "what's the minimum tip?" without a network round-trip.

URI	Description
bolttx://plans	Full plan table — Starter / Growth / Pro / Whale with TPS, min tip, upgrade thresholds.
bolttx://tip-addresses	The 9 canonical BoLt1-9 tip addresses.
bolttx://quickstart	Step-by-step minimum working example. Useful when the agent needs reasoning context for tx construction.

Standard agent workflow

When you ask your AI "send 0.1 SOL to X with minimum tip", here's the sequence the agent follows internally:

1
Call `bolttx_get_my_account` → learn your plan (e.g. Growth)
2
Call `bolttx_get_tip_amount("growth")` → get minimum tip in lamports
3
Call `bolttx_list_tip_addresses` → pick one of the 9 tip addresses
4
Build the transaction locally (your ix + SystemProgram transfer to tip address)
5
Sign the transaction locally with your keypair (never touches the MCP server)
6
Call `bolttx_send_transaction(base64)` → receive signature
7
Optionally call `bolttx_get_status(signature)` after a short wait → confirm landing

Compatibility

Any MCP-compatible client works — the protocol is standardized and the config format is identical across vendors. Per-client installation paths (config file locations) live in the @bolttx/mcp-server README on npm, which we keep up-to-date as new MCP clients ship.

Environment variables

Parameter	Required	Description
BOLTTX_API_KEY	Required	Your BoltTx API key from the dashboard. Required.
BOLTTX_BASE_URL	Optional	Override the API base URL (default: https://bolttx.io). For staging or self-hosted testing only.

Troubleshooting

"bolttx tool is not available"

Confirm the server is listed in your MCP client's active server panel. Check for config parse errors in your MCP client's developer console. Restart the client after editing config.

"Rate limit exceeded"

You've hit your plan's TPS cap. The error body includes `retry_after_ms`. Your AI agent should respect this — or upgrade your plan.

"Transaction must include a tip transfer"

Your transaction is missing the mandatory SystemProgram transfer to a BoltTx tip address. Make sure your AI included `bolttx_list_tip_addresses` and added a transfer ix before submitting.

Libraries Back to Docs Home