imperal-mcp
Enables LLMs to build, validate, smoke test, and deploy Imperal apps using a declarative IR format directly from the editor.
README
imperal-mcp
A stdio MCP server that lets any LLM (Claude Code, Codex, Cursor, etc.) build and deploy Imperal apps using the declarative IR format.
The client's own LLM authors the app; this server validates, smokes, and deploys it — no hand-written deployment scripts needed.
Install & sign in
pipx install imperal-mcp # or: pip install --user imperal-mcp
imperal-mcp login # opens your browser to sign in to Imperal
login stores credentials in ~/.imperal/credentials.json and auto-refreshes the access
token — no manual token management needed.
To sign out and remove stored credentials:
imperal-mcp logout
Configure (.mcp.json)
Add to your project's .mcp.json (or Claude Code's global MCP config):
{
"mcpServers": {
"imperal": {
"command": "imperal-mcp"
}
}
}
No token required — imperal-mcp login provides it.
Environment variables
| Variable | Default | Purpose |
|---|---|---|
IMPERAL_API_URL |
https://auth.imperal.io |
Auth / API base URL |
IMPERAL_PANEL_URL |
https://panel.imperal.io |
Panel base URL |
IMPERAL_TOKEN |
(not set) | Optional override for CI / headless environments — set this to skip browser login |
For CI or headless use, set IMPERAL_TOKEN to a service token from
panel.imperal.io → Developer → Access tokens.
Tools
| Tool | Description |
|---|---|
validate_ir(app_ir) |
Validate an app.ir.json locally — envelope structure + every declarative step. No network. Returns {valid, issues}. |
smoke_ir(app_ir, function, args?) |
Run one function in an isolated store and report {ok, result, trace}. |
deploy_ir(app_ir, app_id) |
Deploy an app.ir.json into the caller's account (creates the app record if needed). |
list_apps() |
List the caller's developer apps (PII-masked). |
get_app(app_id) |
Get one app's manifest + tools with action_type (PII-masked). |
run_read_tool(app_id, function, args?) |
Run a read-only tool of a deployed app. Refuses write/destructive tools. |
Resources
| URI | Description |
|---|---|
imperal://ir-spec |
IR envelope specification — structure, fields, action vocabulary. |
imperal://ui-catalog |
All ui.* component names usable in panels and render steps. |
imperal://examples |
Example app.ir.json (link-saver) to use as a starting point. |
Prompt
build_imperal_app — step-by-step guidance for an LLM to go from intent to a deployed app:
read the spec, author the IR, validate, smoke, deploy.
Security
run_read_tool is gate-kept: it looks up the tool's action_type from get_app and refuses
any tool that is not explicitly action_type: "read". Write and destructive tools are never
executed. All read responses pass through a client-side PII scrub (email/phone redaction) before
being returned to the LLM.
Development
git clone https://github.com/imperalcloud/imperal-mcp
cd imperal-mcp
pip install -e '.[dev]'
python -m pytest
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。