Stata GUI MCP
通过 Windows Stata Automation COM,让 Claude Code、Codex、OpenCode、Claude Desktop、Cursor 和 VS Code 控制可见的 Stata GUI,并把每次执行的真实 text log 自动返回给 AI。
A Windows MCP server that controls a visible Stata GUI through Stata Automation COM and automatically returns the real Stata text log for every execution.
中文
项目特点
- 真实 GUI:命令在可见的 Stata GUI 中运行,而不是隐藏的 Python 模拟环境。
- 人工接管:AI 执行后,用户可以直接在同一个 Stata 窗口中检查、修改和继续分析。
- 持久 Session:同一个
session_id复用同一个 Stata GUI;数据、宏、矩阵、估计结果和工作目录可继续使用。 - 自动返回结果:运行若干行代码或完整 do 文件后,MCP 自动读取本次 text log 并返回给 AI。
- 完整代码块:多行选择通过临时 do 文件整体执行,支持循环、程序定义和续行语法。
- 多任务窗口:不同
session_id可以维护不同的 Stata GUI 和项目日志。 - 项目本地记录:session、do 文件和最新日志关系保存在项目的
.stata-mcp/中。 - 不改原始脚本:运行完整 do 文件时不会重写用户源文件。
- pip 安装:无需下载仓库或在 MCP 配置中填写源码路径。
适用范围
| 要求 | 说明 |
|---|---|
| 操作系统 | Windows 10/11 |
| Stata | 已安装并获得授权的 Stata 17/18/19,MP、SE 或 BE 均可 |
| Automation | Stata COM 接口已注册 |
| Python | 3.10–3.13 |
| MCP 传输 | 本地 stdio |
本项目当前不支持 macOS/Linux,也不使用 PyStata、Stata CLI 或 batch 后端。pip 包只包含 MCP Python 代码,不包含 Stata 软件、许可证或第三方 ado 包。
安装
1. 安装 Python 包
在 PowerShell 中运行:
pip install stata-gui-mcp
pip 会同时安装 MCP Python SDK 和 Windows 所需的 pywin32。用户不需要克隆 GitHub 仓库。
2. 注册 Stata Automation
以管理员身份打开 PowerShell,按实际安装路径运行:
Start-Process -FilePath "C:\Program Files\Stata18\StataMP-64.exe" -ArgumentList "/Register" -Wait
常见可执行文件名包括:
StataMP-64.exe
StataSE-64.exe
StataBE-64.exe
请根据 Stata 版本、edition 和安装位置调整路径。不要直接在 Git Bash 中运行 /Register,因为 Git Bash 可能把它改写成文件路径。
正常启动时,MCP 直接调用已注册的 COM ProgID stata.StataOLEApp,不要求手工填写 Stata 安装路径。
MCP 客户端配置
所有客户端都启动同一个本地 stdio 命令:
stata-gui-mcp
| 客户端 | 推荐方式 | 配置键/文件 |
|---|---|---|
| Claude Code | claude mcp add |
user scope 或 .mcp.json |
| Codex CLI | codex mcp add |
%USERPROFILE%\.codex\config.toml |
| OpenCode | 配置文件或交互式添加 | %USERPROFILE%\.config\opencode\opencode.json |
| Claude Desktop | Desktop 配置 | %APPDATA%\Claude\claude_desktop_config.json |
| Cursor | mcp.json |
%USERPROFILE%\.cursor\mcp.json 或 .cursor\mcp.json |
| VS Code | mcp.json |
用户 MCP 配置或 .vscode\mcp.json |
Claude Code
推荐注册到 user scope,使当前用户的所有项目都能使用:
claude mcp add --transport stdio --scope user stata -- stata-gui-mcp
命令含义:
--transport stdio 使用本地标准输入输出传输
--scope user 当前用户的所有项目可用
stata Claude Code 中显示的 server 名称
-- 分隔 Claude 参数和服务器命令
stata-gui-mcp pip 安装的服务器命令
如果只希望项目内使用,在项目根目录创建 .mcp.json:
{
"mcpServers": {
"stata": {
"type": "stdio",
"command": "stata-gui-mcp",
"args": [],
"env": {}
}
}
}
重新启动 Claude Code 或重新加载 MCP 后,在工具列表中应看到以 mcp__stata__ 开头的工具。
Codex CLI
通过命令添加本地 stdio server:
codex mcp add stata -- stata-gui-mcp
也可以编辑用户配置 %USERPROFILE%\.codex\config.toml,或项目级 .codex\config.toml:
[mcp_servers.stata]
command = "stata-gui-mcp"
args = []
保存配置后重新启动 Codex。codex mcp list 可列出已注册的 server。
OpenCode
OpenCode 可以交互式添加:
opencode mcp add stata
选择 local server,并将命令设为 stata-gui-mcp。也可以编辑用户配置 %USERPROFILE%\.config\opencode\opencode.json,或项目根目录的 opencode.json:
{
"$schema": "https://opencode.ai/config.json",
"mcp": {
"stata": {
"type": "local",
"command": ["stata-gui-mcp"],
"enabled": true
}
}
}
OpenCode 使用 mcp 键,并把本地命令写成数组;它与 Claude Desktop 的 mcpServers 格式不同。保存后重新启动 OpenCode,或运行 opencode mcp list 检查状态。
Claude Desktop
打开以下 Windows 配置文件:
%APPDATA%\Claude\claude_desktop_config.json
加入:
{
"mcpServers": {
"stata": {
"command": "stata-gui-mcp",
"args": []
}
}
}
如果文件中已有其他 server,只添加 stata 项,不要覆盖其他配置。完全退出 Claude Desktop 后重新打开,再在 Developer/MCP 页面检查工具。
Cursor
全局配置文件:
%USERPROFILE%\.cursor\mcp.json
项目配置文件:
<project>\.cursor\mcp.json
内容:
{
"mcpServers": {
"stata": {
"command": "stata-gui-mcp",
"args": []
}
}
}
也可以从 Cursor 的 Settings → Tools & MCP 添加。保存后重新加载 Cursor 窗口并启用 stata server。
VS Code
项目级配置文件:
<project>\.vscode\mcp.json
内容:
{
"servers": {
"stata": {
"type": "stdio",
"command": "stata-gui-mcp",
"args": []
}
}
}
用户级配置可通过命令面板运行 MCP: Open User Configuration 打开。VS Code 使用 servers 键,不是 mcpServers。保存后在 MCP server 列表中启动 stata。
命令不在 PATH 中
如果客户端报告找不到 stata-gui-mcp,可以改用同一 Python 环境的模块入口:
{
"command": "python",
"args": ["-m", "stata_mcp"]
}
Claude Code 对应命令:
claude mcp add --transport stdio --scope user stata -- python -m stata_mcp
使用虚拟环境时,把 python 换成该环境解释器的绝对路径,例如:
{
"command": "D:\\Python\\python.exe",
"args": ["-m", "stata_mcp"]
}
快速开始
运行一个完整 do 文件
stata_run_dofile(
path="D:/research/project/analysis.do",
session_id="main",
role="entry"
)
MCP 会:
- 通过 COM 启动或连接一个可见的 Stata GUI;
- 将工作目录切换到 do 文件目录;
- 覆盖该 session 的最新运行日志;
- 执行原始 do 文件;
- 自动读取并返回本次 Stata 输出。
在同一个 session 中继续
stata_run(commands="ereturn list\npredict double yhat\nsummarize yhat")
多行字符串作为一个临时 do 文件整体执行,执行完成后临时文件会删除。Stata 内存状态继续保留在同一个 GUI 中。
查看数据结构快照
stata_get_data_schema(
sample_rows=20,
include_codebook=true,
include_sample=true,
include_missing=true
)
快照包括:
describe
codebook, compact
misstable summarize
list in 1/20, abbreviate(20)
notes
label dir
Session 与日志
项目运行后创建:
<project>/.stata-mcp/
├── cache/
│ ├── task_registry.json
│ ├── task_registry.json.lock
│ ├── run_<session>_<hash>.log
│ └── schema_<session>_<hash>.log
└── dofiles/
| 文件 | 内容 | 更新方式 |
|---|---|---|
task_registry.json |
session、entry/source/current do 和日志路径 | session 操作时更新 |
task_registry.json.lock |
防止多个 MCP 进程同时更新 registry 时丢失数据 | registry 事务期间自动加锁 |
run_<session>_<hash>.log |
最近一次命令代码块或完整 do 文件输出 | 每次 stata_run/stata_run_dofile 覆盖 |
schema_<session>_<hash>.log |
最近一次数据结构快照 | 每次 stata_get_data_schema 覆盖 |
dofiles/ |
MCP 通过相对文件名生成的 do 文件 | 写文件工具调用时更新 |
日志读取后不会自动删除。运行日志和 schema 日志相互独立。建议把 .stata-mcp/ 加入分析项目自己的 .gitignore。
每个 session 固定使用一个项目缓存目录。不同 session_id 的文件名包含不同哈希,不会互相覆盖。
输出长度
运行工具默认最多把日志末尾 200,000 个字符返回给 AI。超过时,响应会标记 output_truncated: true 并保留完整 log_path。磁盘日志不会被截断。
可选环境变量:
{
"env": {
"STATA_MCP_MAX_OUTPUT_CHARS": "300000"
}
}
工具参考
| 工具 | 用途 |
|---|---|
stata_run |
在最近 session 中将若干行代码作为一个代码块执行,覆盖并返回最新运行日志 |
stata_run_dofile |
在指定 session 中运行完整 do 文件,覆盖并返回最新运行日志 |
stata_session |
list、get、destroy 或 set_recent session |
stata_write_dofile |
写入 do 文件;相对名称写到项目 .stata-mcp/dofiles/ |
stata_read_dofile |
读取 do 文件 |
stata_append_dofile |
向已有 do 文件追加代码 |
stata_read_log |
读取最近运行日志或显式日志路径,支持 full、core、dict |
stata_install_package |
在当前 Stata GUI 中执行 ssc install 或 net install |
stata_get_results |
执行 return list 或 ereturn list 并返回结果 |
stata_get_data_info |
执行 describe 并返回结果 |
stata_get_data_schema |
通过独立 schema 日志返回结构、缺失摘要和样本 |
stata_status |
显示 COM、内存 session 和项目 registry 状态 |
stata_run_dofile 仍接受旧版 log_mode 参数以保持客户端兼容,但 1.1 版始终使用 replace,保证每个 session 只保留最近一次运行输出。
与其他 Stata MCP 的比较
不同项目针对不同工作流,没有一个后端适合所有场景。
| 维度 | stata-gui-mcp(本项目) |
SepineTam/mcp-for-stata | hanlulong/stata-mcp |
|---|---|---|---|
| 主要后端 | Windows COM Automation | Stata CLI/批处理导向 | PyStata worker |
| GUI | 可见 Stata GUI | 通常无 GUI | 通常无 Stata GUI |
| 人工接管 | 可直接接管同一窗口 | 不是主要目标 | 不是主要目标 |
| 状态持续 | 同一 COM GUI session 持续 | 取决于其执行方式 | 持久 PyStata worker |
| AI 获取输出 | 每次自动读取 session text log | CLI/log 输出 | PyStata输出和临时日志 |
| 选中多行代码 | 临时 do 文件整体执行 | 支持命令/文件执行 | 支持 |
| 多 session | 多个 COM GUI | 以对应项目当前实现为准 | 多 worker 进程 |
| 主要平台 | Windows | 跨平台 CLI 场景,具体支持见上游 | PyStata 支持的平台,具体支持见上游 |
| 更适合 | 希望看见、核查并手工继续 Stata | 自动化、批处理和服务器工作流 | 无 GUI 持久会话与深度 IDE 集成 |
本项目的主要优势是可见 GUI + 人工接管 + AI 自动读取真实日志。如果目标是 Linux 服务器、纯批处理或不显示 GUI,应选择 CLI/PyStata 类型方案。
环境变量
普通用户通常无需设置。
| 变量 | 默认值 | 用途 |
|---|---|---|
STATA_COM_PROG_ID |
stata.StataOLEApp |
非标准 COM ProgID |
STATA_EXE |
未设置 | 特殊安装位置需要显式预启动 Stata 时的可执行文件路径;MCP 只保留一个存活的预启动进程,退出后会按需重启,Automation 会话仍由 COM ProgID 建立 |
STATA_MCP_MAX_OUTPUT_CHARS |
200000 |
自动返回给 AI 的日志字符上限 |
STATA_MCP_DIR |
安装包根目录 | 仅用于状态诊断;项目缓存仍跟随 do 文件 |
默认情况下,COM 注册信息负责定位 Stata,不会把 Python site-packages 的上一级误认为 Stata 安装目录。
安全说明
- MCP 获得的权限等同于当前 Windows 用户和 Stata GUI。
- 内置危险命令检查只是一层意外操作防护,不是完整安全沙箱。
stata_write_dofile和stata_append_dofile会修改明确指定的文件。stata_install_package会访问外部源并修改 Stata ado 环境。- 建议在运行前审阅 AI 生成的分析代码和文件路径。
- MCP 使用具名日志
__stata_mcp_run和__stata_mcp_schema,不会主动执行log close _all。 - 如果用户 do 文件自身执行
log close _all,它也会关闭 MCP 日志,后续输出可能无法捕获;建议只关闭用户自己命名的日志。
故障排查
COM 未注册或无法启动 Stata
以管理员 PowerShell 重新执行 /Register,然后完全退出旧 Stata 进程并重启 MCP 客户端。
找不到 stata-gui-mcp
客户端启动环境的 PATH 与安装 pip 包时的终端可能不同。使用上文的 python -m stata_mcp 配置,或填入正确 Python 解释器绝对路径。
stata_run 提示没有 session
先运行一个绝对路径 do 文件:
stata_run_dofile(path="D:/research/project/analysis.do", session_id="main")
后续无路径命令会发送到最近 session。也可以使用:
stata_session(action="set_recent", session_id="main")
Stata GUI 被关闭
MCP 检测到 COM 断开时会保留旧日志和 registry,重新初始化一个空 GUI,并明确提示内存状态已丢失。中断命令不会自动重放。
AI 只看到日志末尾
检查响应中的 output_truncated 和 log_path,再调用 stata_read_log 读取完整文件,或调整 STATA_MCP_MAX_OUTPUT_CHARS。
开发
源码仓库:https://github.com/shichengg/stata-mcp
git clone https://github.com/shichengg/stata-mcp
cd stata-mcp
python -m venv .venv
.\.venv\Scripts\python -m pip install -e ".[dev]"
.\.venv\Scripts\python -m pytest
欢迎提交 issue 和 pull request。
License
English
Overview
stata-gui-mcp is a Windows MCP server that controls a visible Stata GUI through Stata Automation COM. It keeps Stata state alive per session and automatically returns the actual text log from every selected-code or do-file execution.
Highlights
- Visible Stata GUI with direct manual takeover.
- Persistent Stata data, macros, matrices, estimates, and working directory within a session.
- Multi-line code is executed as one temporary do-file, not line by line.
- One replace-only latest run log per session, returned automatically to the AI.
- A separate replace-only schema snapshot log.
- Multiple
session_idvalues can maintain independent Stata GUI windows. - Original user do-files are never rewritten.
- Installable from pip without cloning the repository.
Requirements
- Windows 10/11.
- A licensed Stata 17/18/19 installation (MP, SE, or BE).
- Stata Automation COM registered with
/Register. - Python 3.10–3.13.
The package does not include Stata or a Stata license. It does not currently provide macOS/Linux, PyStata, CLI, or batch backends.
Installation
Install the package in PowerShell:
pip install stata-gui-mcp
Register the executable that matches your installation from an elevated PowerShell:
Start-Process -FilePath "C:\Program Files\Stata18\StataMP-64.exe" -ArgumentList "/Register" -Wait
Adjust the Stata version, edition, and path as needed. Normal MCP startup locates Stata through the registered COM ProgID; no source path or default STATA_EXE is required.
MCP client setup
All clients start the same local stdio command: stata-gui-mcp.
Claude Code
Claude Code is listed first because it is the primary supported setup:
claude mcp add --transport stdio --scope user stata -- stata-gui-mcp
Project-scoped .mcp.json:
{
"mcpServers": {
"stata": {
"type": "stdio",
"command": "stata-gui-mcp",
"args": [],
"env": {}
}
}
}
Restart Claude Code or reload MCP servers and look for tools prefixed with mcp__stata__.
Codex CLI
codex mcp add stata -- stata-gui-mcp
User configuration %USERPROFILE%\.codex\config.toml or project .codex\config.toml:
[mcp_servers.stata]
command = "stata-gui-mcp"
args = []
Restart Codex after saving. codex mcp list shows configured servers.
OpenCode
Interactive setup:
opencode mcp add stata
Select a local server and enter stata-gui-mcp, or edit %USERPROFILE%\.config\opencode\opencode.json / project opencode.json:
{
"$schema": "https://opencode.ai/config.json",
"mcp": {
"stata": {
"type": "local",
"command": ["stata-gui-mcp"],
"enabled": true
}
}
}
Restart OpenCode or use opencode mcp list.
Claude Desktop
Edit %APPDATA%\Claude\claude_desktop_config.json:
{
"mcpServers": {
"stata": {
"command": "stata-gui-mcp",
"args": []
}
}
}
Merge the stata entry with existing servers, fully quit Claude Desktop, and reopen it.
Cursor
Use global %USERPROFILE%\.cursor\mcp.json or project .cursor\mcp.json:
{
"mcpServers": {
"stata": {
"command": "stata-gui-mcp",
"args": []
}
}
}
Alternatively use Settings → Tools & MCP, then reload the Cursor window.
VS Code
Create project .vscode\mcp.json:
{
"servers": {
"stata": {
"type": "stdio",
"command": "stata-gui-mcp",
"args": []
}
}
}
For a user-level configuration, run MCP: Open User Configuration from the Command Palette. VS Code uses servers, not mcpServers.
PATH fallback
If a client cannot find the console command, use the Python interpreter that installed the package:
{
"command": "python",
"args": ["-m", "stata_mcp"]
}
Claude Code equivalent:
claude mcp add --transport stdio --scope user stata -- python -m stata_mcp
Replace python with the virtual-environment interpreter's absolute path when applicable.
Quick start
Start a persistent GUI session by running a do-file:
stata_run_dofile(
path="D:/research/project/analysis.do",
session_id="main",
role="entry"
)
Continue in the same in-memory Stata state:
stata_run(commands="ereturn list\npredict double yhat\nsummarize yhat")
Capture the current dataset schema:
stata_get_data_schema(sample_rows=20)
Sessions and logs
<project>/.stata-mcp/
├── cache/
│ ├── task_registry.json
│ ├── task_registry.json.lock
│ ├── run_<session>_<hash>.log
│ └── schema_<session>_<hash>.log
└── dofiles/
stata_runandstata_run_dofilereplacerun_<session>_<hash>.logand return it automatically.stata_get_data_schemaseparately replacesschema_<session>_<hash>.log.task_registry.json.lockserializes registry updates across separate MCP processes.- Reading either log does not delete it.
- Different session IDs use distinct hashed names.
- Add
.stata-mcp/to the analysis project's.gitignore.
The automatic response includes at most the last 200,000 characters by default. If output_truncated: true appears, the complete file remains at log_path. Override the limit with STATA_MCP_MAX_OUTPUT_CHARS.
Tool reference
| Tool | Purpose |
|---|---|
stata_run |
Execute a complete code block in the recent session and return its latest run log |
stata_run_dofile |
Run an original do-file in a session and return its latest run log |
stata_session |
List, inspect, destroy, or select the recent session |
stata_write_dofile |
Write a do-file |
stata_read_dofile |
Read a do-file |
stata_append_dofile |
Append Stata code to a do-file |
stata_read_log |
Read a log as full text, core text, or parsed command/result JSON |
stata_install_package |
Run ssc install or net install in the GUI |
stata_get_results |
Return return list or ereturn list output |
stata_get_data_info |
Return describe output |
stata_get_data_schema |
Return schema, missing summary, and sample through a separate log |
stata_status |
Show COM, in-memory session, and registry status |
log_mode remains accepted by stata_run_dofile for old clients, but version 1.1 always replaces the latest session run log.
Comparison
| Dimension | stata-gui-mcp |
SepineTam/mcp-for-stata | hanlulong/stata-mcp |
|---|---|---|---|
| Main backend | Windows COM Automation | Stata CLI/batch-oriented | PyStata workers |
| Visible Stata GUI | Yes | Usually no | Usually no |
| Manual takeover | Same GUI window | Not the primary goal | Not the primary goal |
| Persistent state | Persistent COM GUI session | Depends on execution mode | Persistent PyStata worker |
| AI output | Automatically returned session text log | CLI/log output | PyStata output and temporary logs |
| Multi-session | Multiple COM GUI sessions | See upstream implementation | Multiple worker processes |
| Primary fit | Visible, reviewable, human-in-the-loop Stata | Automation and batch workflows | Headless persistent IDE integration |
The differentiator is visible GUI + manual takeover + automatic real-log return. Choose a CLI/PyStata project when a server, headless, or cross-platform workflow matters more.
Environment variables
| Variable | Default | Purpose |
|---|---|---|
STATA_COM_PROG_ID |
stata.StataOLEApp |
Override the registered COM ProgID |
STATA_EXE |
unset | Explicitly prelaunch a non-standard Stata executable; MCP tracks one live prelaunched process and relaunches it after exit, while Automation sessions are still created through the COM ProgID |
STATA_MCP_MAX_OUTPUT_CHARS |
200000 |
Maximum log characters returned automatically |
STATA_MCP_DIR |
package root | Status diagnostics only |
Security and limitations
- The server runs with the current Windows user's permissions.
- The dangerous-command check is a guardrail, not a sandbox.
- File-writing tools modify explicitly selected files.
- Package installation changes the Stata ado environment and may use the network.
- Review generated analysis code and paths before execution.
- MCP uses named logs and never intentionally issues
log close _all. - A user do-file containing
log close _allcan still close the MCP log and truncate capture; close only user-named logs instead.
Troubleshooting
- COM startup fails: rerun
/Registerfrom elevated PowerShell and restart Stata/MCP clients. - Command not found: use the
python -m stata_mcpfallback with the correct interpreter. - No current session: call
stata_run_dofilewith an absolute path first. - GUI was closed: MCP preserves old logs/registry and reports that the newly initialized GUI has empty memory; it never silently replays the interrupted command.
- Output was truncated: read
log_pathwithstata_read_logor raiseSTATA_MCP_MAX_OUTPUT_CHARS.
Development
git clone https://github.com/shichengg/stata-mcp
cd stata-mcp
python -m venv .venv
.\.venv\Scripts\python -m pip install -e ".[dev]"
.\.venv\Scripts\python -m pytest
Repository: https://github.com/shichengg/stata-mcp
License
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Filter files by name, interpreter, ABI, and platform.
If you're not sure about the file name format, learn more about wheel file names.
Copy a direct link to the current filters
File details
Details for the file stata_gui_mcp-1.1.0.tar.gz.
File metadata
- Download URL: stata_gui_mcp-1.1.0.tar.gz
- Upload date:
- Size: 40.7 kB
- Tags: Source
- Uploaded using Trusted Publishing? No
- Uploaded via:
twine/6.2.0 CPython/3.13.7
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
40c135bfe11f4b3c328243b66c3be9b9f12fdd8aa5096880871992cc614dc1c0
|
|
| MD5 |
e6537d2b09879d5de69259d08b29786f
|
|
| BLAKE2b-256 |
0a4922a2f82c8611183b2c91046cdc916fdb2e0a30bce43504952629992ccf13
|
File details
Details for the file stata_gui_mcp-1.1.0-py3-none-any.whl.
File metadata
- Download URL: stata_gui_mcp-1.1.0-py3-none-any.whl
- Upload date:
- Size: 24.7 kB
- Tags: Python 3
- Uploaded using Trusted Publishing? No
- Uploaded via:
twine/6.2.0 CPython/3.13.7
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
f5ec467dc2ed99ba97b075f10eef970b1f384693abe211c4b80df7e2f0e9d0e2
|
|
| MD5 |
667f26e23bd5cb192e1e8dd5135fe3b1
|
|
| BLAKE2b-256 |
ef4bc6fe6e22daf242385832723b0c9ca4a1ad2c25be5016e6bb619e6455fd44
|