Newcontext-mode—Save 98% of your AI coding agent's context windowLearn more
MCP Directory
ServersClientsBlog

context-mode

Save 98% of your AI coding agent's context window. Works with Claude Code, Cursor, Copilot, Codex, and more.

Try context-mode
MCP Directory

Model Context Protocol Directory

MKSF LTD
Suite 8805 5 Brayford Square
London, E1 0SG

MCP Directory

  • About
  • Blog
  • Documentation
  • Contact

Menu

  • Servers
  • Clients

© 2026 model-context-protocol.com

The Model Context Protocol (MCP) is an open standard for AI model communication.
Powered by Mert KoseogluSoftware Forge
  1. Home
  2. Servers
  3. xlmtools

xlmtools

GitHub

Stellar-native MCP server with 21 pay-per-call tools for AI agents. MCP + CLI + Agent Skill.

0
0
<p align="center"> <img src="./assets/xlmtools-header.svg" alt="XLMTools — Stellar-native MCP server with pay-per-call tools for AI agents" width="100%"/> </p>

XLMTools

Stellar-native MCP server with pay-per-call tools for AI agents. One-line install. No subscriptions. Every payment settles on-chain.

Built for the Stellar Hacks: Agents hackathon (x402 / MPP track).

Quick start

XLMTools ships in three flavours: MCP server, standalone CLI, and cross-client Agent Skill. Install any or all.

MCP server

# Claude Code
claude mcp add xlmtools npx @xlmtools/mcp

# Gemini CLI
gemini mcp add xlmtools npx -y @xlmtools/mcp

# OpenAI Codex
codex mcp add xlmtools npx -y @xlmtools/mcp

# Cursor / Windsurf / Claude Desktop / VS Code / Zed / Cline / Continue / Goose
# Add to the client's MCP config:
#   { "command": "npx", "args": ["-y", "@xlmtools/mcp"] }

See the MCP Host Setup guide for exact config for each of 12 clients.

Standalone CLI

npm install -g @xlmtools/cli
xlm --help

Agent Skill

# Paste into any agent
Read https://xlmtools.com/skill.md and follow the instructions to install XLMTools.

# Or via CLI
npx skills add github:Blockchain-Oracle/xlmtools --skill xlmtools

The skill works natively in Claude Code, Cursor 2.4+, Windsurf, VS Code Copilot, Codex CLI, Gemini CLI, Goose, and Cline.

On first run, XLMTools generates a Stellar testnet wallet, funds it with XLM via friendbot, and adds a USDC trustline — all automatically. This only happens on testnet. The only manual step is getting testnet USDC:

  1. Run the install command above and make any tool call to trigger wallet setup
  2. Go to faucet.circle.com, select Stellar, paste your wallet address
  3. Done — all paid tools will now work

Your wallet is at ~/.xlmtools/config.json. The secret key never leaves your machine.

Note: Auto-wallet funding (friendbot XLM + USDC trustline) only runs on testnet. On mainnet, fund your wallet manually.

What this does

XLMTools gives AI agents access to 21 tools. Paid tools cost $0.001 to $0.04 per call in USDC via Stellar's Micropayment Protocol. Free tools have no cost. No API keys needed. No accounts to create.

Works with every major agent client (Claude Code, Claude Desktop, Cursor, Windsurf, VS Code Copilot, Gemini CLI, OpenAI Codex, Zed, Continue, Cline, Goose) and as a standalone xlm for direct terminal use.

Tools

Paid (USDC via Stellar MPP)

ToolPriceWhat it does
search$0.003Web and news search
research$0.010Multi-source deep research with summaries
youtube$0.002Video search and lookup
screenshot$0.010Capture a screenshot of any URL
scrape$0.002Extract clean text from any URL
image$0.040AI image generation from text prompts
stocks$0.001Real-time stock quotes

Free

ToolWhat it does
cryptoCryptocurrency prices and market data
weatherCurrent weather for any city
domainDomain availability check
budgetSet/check/clear session spending limits
walletYour Stellar wallet address and balance
toolsList all available tools and prices
dex-orderbookStellar DEX live orderbook
dex-candlesOHLCV candlestick data
dex-tradesRecent DEX trade history
swap-quoteBest swap path between assets
stellar-assetAsset info, supply, trustlines
stellar-accountAccount balances and signers
stellar-poolsLiquidity pool data
oracle-priceReflector oracle prices

Agent Skill

XLMTools also ships as an Agent Skill — procedural instructions that teach Claude (or any skill-capable agent) when and how to use XLMTools tools. Three install methods:

# 1. Via the community skills CLI
pnpm dlx skills add github:Blockchain-Oracle/xlmtools --skill xlmtools

# 2. Via prompt — paste into any agent
Read https://xlmtools.com/skill.md and follow the instructions to install XLMTools.

# 3. Manual — copy the SKILL.md from packages/skills/xlmtools/ into ~/.claude/skills/xlmtools/

The skill teaches your agent the decision tree (user intent → tool), paid tool warnings, receipt handling, and CLI fallback when MCP isn't available. See packages/skills/xlmtools/SKILL.md or visit /skill on the frontend.

Standalone CLI

XLMTools also ships as a terminal tool. Same wallet, same payments, no MCP host needed.

npm install -g @xlmtools/cli

xlm wallet
xlm crypto bitcoin,stellar
xlm weather Lagos
xlm search "Stellar MPP" --count 5
xlm dex-orderbook XLM/USDC --limit 3
xlm --help

Every paid call prints a Stellar transaction hash you can verify on-chain.

How payment works

You ask Claude to search for something
  -> CLI sends request to XLMTools API
  -> API returns 402 Payment Required
  -> mppx auto-builds a Soroban USDC transfer on Stellar
  -> Your local key signs it (key never leaves your machine)
  -> API verifies payment, calls the backend, returns results
  -> Response includes a Stellar transaction hash for verification

No manual steps. Every paid response includes a receipt:

Payment: $0.003 USDC · tx/8f3a1b2c... · stellar testnet

Verify any transaction at stellar.expert.

Cost management

Budget — set a per-session spending cap so agents cannot overspend:

Set my budget to $2.00

Calls that would exceed the limit are blocked. Use budget check to see remaining balance.

Caching — identical queries within 5 minutes return cached results at no charge. You never pay twice for the same data.

Architecture

Claude / Cursor / Windsurf / VS Code Copilot / …    (MCP host)
    |
    | stdio (when user installs via claude mcp add xlmtools npx @xlmtools/mcp)
    v
@xlmtools/mcp                                       (thin wrapper, ~15 lines)
    |  imports createMcpServer() factory from @xlmtools/cli
    |  connects StdioServerTransport, done
    v
@xlmtools/cli  (also runnable as standalone `xlm` via Bash)
    |  - auto-generated Stellar wallet at ~/.xlmtools/config.json
    |  - mppx polyfill handles 402 payments transparently
    |  - X-XLMTools-Client header → per-address stats attribution
    |  - session budget tracking + 5-min response caching
    |
    | HTTPS + signed MPP credentials
    v
XLMTools API Server (Express, hosted)
    |  - verifies payments via Soroban simulation
    |  - calls backend APIs (Brave, Exa, OpenAI, etc.)
    |  - users never need backend API keys
    |  - in-memory call log for /stats + /stats/by-client
    |
    v
Stellar Testnet (USDC settlement via Soroban SAC)

Two install paths, one source of truth

XLMTools ships as two packages instead of one to make the install story work cleanly across every MCP host:

PackageBinWhat you installPurpose
@xlmtools/mcpxlmtools-mcpclaude mcp add xlmtools npx @xlmtools/mcpMCP stdio server. Thin wrapper that imports the server factory from @xlmtools/cli. Single bin, so npx @xlmtools/mcp auto-resolves.
@xlmtools/clixlmnpm install -g @xlmtools/cliStandalone terminal CLI. Also exports createMcpServer() via main/types — consumed by @xlmtools/mcp as a runtime dependency.

Why two packages? npm/npx can't auto-resolve npx @pkg/name when a package has multiple bins and none matches the package tail. The ecosystem convention (@upstash/context7-mcp, @magicuidesign/mcp, @modelcontextprotocol/server-*) is one package per bin. Shipping the MCP as its own single-bin package is the only clean way to make claude mcp add xlmtools npx @xlmtools/mcp work in every host, while still giving terminal users npm install -g @xlmtools/cli for the xlm command.

Project structure

packages/
  mcp/     - @xlmtools/mcp — thin stdio wrapper, published to npm
  cli/     - @xlmtools/cli — standalone `xlm` + createMcpServer() factory,
             owns all tool implementations + shared lib code, published to npm
  api/     - Express API (hosted at api.xlmtools.com)
             verifies MPP, calls backends, per-address stats
  web/     - Next.js 16 frontend (xlmtools.com)
             landing page, tools browser, Explorer live stream, Stats history
  docs/    - Nextra 4 documentation site (docs.xlmtools.com)
  skills/  - Cross-client Agent Skill (SKILL.md), distributed via skills CLI

Documentation

Full documentation covering installation, all 21 tools, payment flow, MCP host setup for Claude/Cursor/Windsurf, API reference, and FAQ.

cd packages/docs && pnpm dev

Docker

# 1. Copy and configure env
cp packages/api/.env.example packages/api/.env
# Edit packages/api/.env with your API keys

# 2. Run
docker compose up -d

# 3. Verify
curl http://localhost:3000/health

The Docker setup runs the API server only. Both the MCP (@xlmtools/mcp) and standalone CLI (@xlmtools/cli) run locally on the user's machine — spawned via npx or installed globally via npm install -g. They hit the hosted API over HTTPS.

Development

pnpm install

# API server (needs .env with backend API keys)
pnpm dev:api

# CLI in dev mode
pnpm dev:cli

# Frontend
cd packages/web && pnpm dev

# Docs
cd packages/docs && pnpm dev

Environment variables (API server only)

VariablePurpose
STELLAR_RECIPIENTStellar public key receiving USDC payments
MPP_SECRET_KEYRandom string for mppx credential verification
BRAVE_API_KEYBrave Search API
EXA_API_KEYExa research API
YOUTUBE_API_KEYYouTube Data API
SCREENSHOTONE_KEYScreenshotOne API
OPENAI_API_KEYOpenAI (DALL-E 3)
ALPHA_VANTAGE_KEYAlpha Vantage stock data
OPENWEATHER_API_KEYOpenWeatherMap

Users never configure these. They install, fund their wallet, and use tools.

Tech stack

  • TypeScript, Node.js 22+, ES modules
  • @modelcontextprotocol/sdk (MCP)
  • @stellar/stellar-sdk, @stellar/mpp, mppx (Stellar MPP payments)
  • Express 5 (API), Next.js 16 (frontend), Nextra 4 (docs)
  • Zod 4 (validation), pino (logging)
  • Stellar testnet, USDC

What makes this different

  1. First MCP server on Stellar with MPP billing
  2. Pay per call, not per month. No API keys for users to manage
  3. 21 tools covering web, media, finance, and the full Stellar DEX
  4. Response caching and budget enforcement built in
  5. Every payment verifiable on-chain with a real Stellar transaction hash
  6. Direct settlement via MPP charge mode on Soroban

License

MIT

Repository

BL
Blockchain-Oracle

Blockchain-Oracle/xlmtools

Created

April 11, 2026

Updated

April 13, 2026

Language

TypeScript

Category

AI