Prism MCP Server
Enables AI assistants to interact with InterSystems IRIS databases through SQL queries, document management, ObjectScript execution, debugging, and unit testing via the Atelier REST API.
README
<div align="center"> <img src="logo.svg" width="200" alt="Prism Logo" />
Prism
Prism lets AI see through IRIS.
MCP server and CLI for InterSystems IRIS development — SQL queries, document management, compilation, debugging, testing, and ObjectScript execution via the Atelier REST API.
Features
- SQL — Run queries, DDL, stored procedures against any IRIS namespace
- Documents — Upload, fetch, compile, and delete
.cls,.mac,.incfiles - Terminal — Execute ObjectScript via native (SuperServer) or WebSocket backend
- Debugging — Interactive step-through debugger with breakpoints, variable inspection, and stack traces
- Testing — Run
%UnitTesttest classes, list test methods, view historical results - MCP Server — Expose all tools to AI assistants (Claude Code, Claude Desktop, Cursor, GitHub Copilot)
- Cast Plugins — Extend Prism with custom commands from any Git repository
- Cross-platform — Windows installer, Linux/macOS via pip/uv
Quick Start
# Install
uv sync
# Configure (or use environment variables)
uv run prism config -u _SYSTEM -p SYS -U http://localhost:52773 -n USER
# Run a SQL query
uv run prism sql "SELECT TOP 5 Name FROM %Dictionary.ClassDefinition"
# Start the MCP server
uv run prism serve
Need an IRIS instance? Start one with Docker:
docker run -d --name iris -p 52773:52773 -p 1972:1972 intersystemsdc/iris-community:latest
CLI Commands
| Command | Description |
|---|---|
prism sql |
Run an SQL query |
prism terminal |
Run ObjectScript via native SuperServer |
prism ws |
Run ObjectScript via WebSocket |
prism put-doc |
Upload a file to IRIS |
prism get-doc |
Fetch a document from IRIS |
prism list-docs |
List source documents |
prism delete-doc |
Delete a document |
prism compile |
Compile documents |
prism info |
Server version and namespaces |
prism test |
Run unit test classes |
prism list-tests |
Discover test classes |
prism config |
View or edit settings |
prism cast |
Run custom commands from Git repos |
prism serve |
Start the MCP server |
Global option: prism --format toon for TOON output.
Shell completion
Tab completion is available for bash, zsh, fish, and PowerShell:
prism --install-completion # auto-detect shell
After running, restart your terminal. Then prism conf + Tab auto-completes
to prism config, prism s + Tab cycles through sql and serve, etc.
See the commands overview for details.
MCP Tools
10 tools are always available, 2 workspace-gated (put_document, put_and_compile),
and 9 debug-gated (debug_*) — up to 21 total.
See the full tool reference for details.
MCP Client Configuration
Claude Code
Add to ~/.claude/settings.json or .claude/settings.json:
{
"mcpServers": {
"iris": {
"url": "http://localhost:3000/mcp"
}
}
}
Claude Desktop
Add to claude_desktop_config.json:
{
"mcpServers": {
"iris": {
"command": "uv",
"args": ["run", "--directory", "/path/to/prism", "prism", "serve"]
}
}
}
GitHub Copilot (VS Code)
Add to .vscode/mcp.json:
{
"servers": {
"iris": {
"type": "http",
"url": "http://localhost:3000/mcp"
}
}
}
Cursor
Add to .cursor/mcp.json:
{
"mcpServers": {
"iris": {
"url": "http://localhost:3000/mcp"
}
}
}
Environment Variables
| Variable | Default | Description |
|---|---|---|
IRIS_BASE_URL |
http://localhost:52773 |
IRIS instance URL |
IRIS_USERNAME |
_SYSTEM |
Authentication username |
IRIS_PASSWORD |
SYS |
Authentication password |
IRIS_NAMESPACE |
USER |
Default namespace |
IRIS_WORKSPACE |
(empty) | Local directory for MCP file I/O tools |
IRIS_COMPILE_FLAGS |
cuk |
Compiler flags |
IRIS_DEBUG_ENABLED |
false |
Enable debug tools (debug_*) |
IRIS_TERMINAL_METHOD |
native |
Terminal backend: native or ws |
See the configuration guide for all 21 settings.
Project Structure
src/prism/
├── settings.py # Pydantic settings (env, .env, config.json)
├── iris/
│ ├── sdk/ # HTTP client, workspace, debug protocols, terminal
│ └── api/ # Thin IRIS REST API wrappers (sql, docs, compile, debug)
├── mcp/ # MCP tools with @logged_tool decorator
│ ├── _decorator.py # Logging + auto-discovery
│ ├── server.py # FastMCP server
│ └── *.py # One module per tool domain
├── cast/ # Cast plugin system (import-based Typer plugins)
│ └── manager.py # Clone, import, cache, run commands
└── cli/ # Typer CLI commands (async wrappers)
Testing
uv run pytest tests/unit/ -v # No IRIS needed (276 tests)
IRIS_BASE_URL=http://localhost:52773 \
uv run pytest tests/integration/ -v # Needs IRIS (72 tests)
uv run ruff check . && uv run ruff format --check . # Lint
Full testing guide: docs/testing.md
Releases
Prism follows a Git Flow
release workflow with two protected branches: development (active work)
and main (stable releases).
Download the latest Windows installer or standalone exe from GitHub Releases.
| Artifact | Description |
|---|---|
prism-X.Y.Z-setup.exe |
Windows installer (Inno Setup, adds to PATH) |
prism.exe |
Standalone Windows binary (PyInstaller) |
prism-X.Y.Z-py3-none-any.whl |
Python wheel (pip install prism) |
See the release guide for the full release workflow, branch model, hotfix procedure, and CI pipeline details.
Documentation
Full documentation at adriaerni.github.io/Prism
License
Copyright © 2026 Adria Sanchez.
Licensed under the GNU Affero General Public License v3.0 (AGPL-3.0). You may use, modify, and distribute this software freely, including for commercial purposes, as long as you share your source code under the same license. SaaS/hosting providers must also share their source code (network clause).
See LICENSE for full terms.
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。