IrishTaxHub MCP Server
Dynamic MCP server that exposes IrishTaxHub API operations as tools via OpenAPI-powered discovery and invocation, allowing users to list endpoints, get request schemas, and call any API endpoint dynamically.
README
IrishTaxHub MCP Server
Model Context Protocol (MCP) server that exposes IrishTaxHub API operations as tools.
This server now uses FastMCP v2 for a simpler, production-ready implementation.
Features
- Dynamic OpenAPI-powered tools only:
openapi_list_endpoints(tag?)→ discover available endpointsopenapi_get_request_schema(path, method=POST)→ JSON Schema for request bodyopenapi_invoke(path, method=POST, body?, params?)→ validate and call dynamically
Install
Use Poetry (recommended) or pip.
# With Poetry
poetry install
# Or with pip
pip install -e .
# If running directly with FastMCP CLI
pip install fastmcp
Configure
Set environment variables (or copy .env.example to .env):
For production use:
export IRISHTAXHUB_BASE_URL="https://prod.aws.irishtaxhub.ie"
export IRISHTAXHUB_TIMEOUT="30"
# OpenAPI source (file or URL). Defaults to "$IRISHTAXHUB_BASE_URL/apispec_1.json".
export IRISHTAXHUB_OPENAPI="https://prod.aws.irishtaxhub.ie/apispec_1.json"
export IRISHTAXHUB_DEVELOPMENT_MODE=false
For local development (if running the API locally):
export IRISHTAXHUB_BASE_URL="http://localhost:5000"
export IRISHTAXHUB_TIMEOUT="30"
export IRISHTAXHUB_OPENAPI="http://localhost:5000/openapi.json"
export IRISHTAXHUB_DEVELOPMENT_MODE=true
Run (local)
Standard I/O server (used by MCP clients):
python -m irishtaxhub_mcp.server
# or via Poetry
poetry run irishtaxhub-mcp
# or run with the FastMCP CLI (equivalent)
fastmcp run src/irishtaxhub_mcp/server.py
Use with Claude Desktop
Add an MCP server entry to your Claude Desktop config (replace path as needed):
For production use:
{
"mcpServers": {
"irishtaxhub": {
"command": "python",
"args": ["-m", "irishtaxhub_mcp.server"],
"env": {
"IRISHTAXHUB_BASE_URL": "https://prod.aws.irishtaxhub.ie",
"IRISHTAXHUB_OPENAPI": "https://prod.aws.irishtaxhub.ie/apispec_1.json"
}
}
}
}
For local development:
{
"mcpServers": {
"irishtaxhub": {
"command": "python",
"args": ["-m", "irishtaxhub_mcp.server"],
"env": {
"IRISHTAXHUB_BASE_URL": "http://localhost:5000",
"IRISHTAXHUB_OPENAPI": "http://localhost:5000/openapi.json"
}
}
}
}
Then restart Claude Desktop; tools will appear as openapi_list_endpoints,
openapi_get_request_schema, and openapi_invoke.
Examples
Dynamic discovery + invocation:
{
"tool": "openapi_list_endpoints",
"arguments": { "tag": "Tax Calculators" }
}
{
"tool": "openapi_get_request_schema",
"arguments": { "path": "/v1/tax/calculators/refund", "method": "POST" }
}
{
"tool": "openapi_invoke",
"arguments": {
"path": "/v1/tax/calculators/refund",
"method": "POST",
"body": {
"marital_status": "single",
"year": 2024,
"employment_income": { "income": 50000 }
}
}
}
Notes
- Tools call your deployed HTTP API; they do not import internal Python modules. This keeps the MCP server lightweight and decoupled.
- Dynamic tools load your OpenAPI definition from a file or via HTTP. For local development, you can point
IRISHTAXHUB_OPENAPIat your repo’sopenapi.yaml— Jinja templating is supported. - If you prefer in-process usage, you can create another variant inside the API repo that imports facades directly.
Claude PR Review (GitHub Action)
This repo includes a GitHub Action (.github/workflows/claude-pr-review.yml) that automatically reviews PRs using Claude. It requires a CLAUDE_CODE_OAUTH_TOKEN secret.
Generating the token
- Run
claude setup-tokenin your terminal — this opens a browser to authenticate with your Claude account and outputs an OAuth token - Copy the token
Setting the secret
gh secret set CLAUDE_CODE_OAUTH_TOKEN --repo jhurleyai/irishtaxhub-mcp
Paste the token when prompted.
When to regenerate
The token is tied to your Claude subscription. If your subscription lapses and you renew, the existing token is invalidated — you must run claude setup-token again and update the secret.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。