stata-mcp
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@stata-mcp Run do-file analysis.do and show the regression output."
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Stata MCP Server
通过 Windows Stata Automation COM 让 Claude Code 控制 Stata GUI 的 MCP 服务器。
A Claude Code MCP server that controls the Stata GUI through Windows Stata Automation COM.
中文
功能
通过 Stata Automation COM 在真实 Stata GUI 窗口中运行命令。
按 do 文件绝对路径维护 Stata 会话:同一个 do 文件复用同一个 Stata 窗口。
不同 do 文件打开不同 Stata 窗口,方便同时浏览多个分析任务。
不带路径的命令会发送到最近一次运行的 do 文件会话。
支持 do 文件写入、读取、追加和运行。
支持读取 Stata text log,供 Claude 分析输出结果。
支持读取当前数据结构、缺失情况和样本预览,辅助 Claude 修改 do 文件。
演示效果

目录结构
D:/Stata18/mcp/
├── README.md
├── pyproject.toml
├── .gitignore
├── stata_mcp.py # 兼容启动器
├── src/
│ └── stata_mcp/
│ ├── __init__.py
│ └── server.py # MCP 服务器主文件
├── runtime/
│ ├── dofiles/ # Claude/MCP 默认生成 do 文件
│ └── logs/ # Claude/MCP 默认读取或生成 log
└── examples/ # 示例 do 文件环境要求
Windows
Stata 18 MP,已注册 Automation COM
Python 3.10+
Python 包:
mcp,pywin32Claude Code
推荐安装位置
推荐把本项目放在 Stata 安装目录下的 mcp 文件夹,例如:
D:/Stata18/mcp如果 Stata 安装在其他位置,也建议放在对应目录下,例如:
C:/Program Files/Stata18/mcp高级用户也可以放在任意稳定目录。无论放在哪里,Claude Code MCP 配置中的脚本路径都必须指向实际的 stata_mcp.py。
安装步骤
下载或 clone 本仓库到 Stata 安装目录下的
mcp文件夹。安装 Python 依赖:
pip install mcp pywin32也可以在虚拟环境中安装:
python -m venv .venv
.\.venv\Scripts\Activate.ps1
pip install mcp pywin32以管理员身份打开 PowerShell,注册 Stata Automation。按你的 Stata 安装路径修改命令:
Start-Process -FilePath "D:\Stata18\StataMP-64.exe" -ArgumentList "/Register" -Wait验证 COM 是否可用:
python -c "import win32com.client; s=win32com.client.Dispatch('stata.StataOLEApp'); s.DoCommand('display 12345')"预期结果:Stata GUI 打开或被连接,并显示 12345。
添加到 Claude Code(推荐)
推荐使用 claude mcp add:
claude mcp add stata -- python D:\Stata18\mcp\stata_mcp.py如果安装在其他路径,请替换为实际路径。例如:
claude mcp add stata -- python "C:\Program Files\Stata18\mcp\stata_mcp.py"添加后重启 Claude Code,确认 MCP 工具列表中出现 stata 工具。
手动配置 Claude Code MCP
如果不使用 claude mcp add,也可以手动配置 MCP JSON:
"stata": {
"command": "python",
"args": ["D:\\Stata18\\mcp\\stata_mcp.py"]
}也可以直接指向主文件:
"stata": {
"command": "python",
"args": ["D:\\Stata18\\mcp\\src\\stata_mcp\\server.py"]
}通常建议使用兼容启动器 stata_mcp.py,这样后续项目内部结构调整时配置更稳定。
可配置环境变量
一般用户不需要设置。如果你的安装路径或 COM 名称特殊,可以设置:
变量 | 默认值 | 作用 |
| 自动识别项目根目录 | MCP 项目目录 |
| MCP 上一级目录 | Stata 安装目录 |
|
| README 注册提示用 |
|
| Stata Automation ProgID |
工具
工具 | 功能 |
| 在最近 do 文件会话中执行命令 |
| 按 do 文件绝对路径运行/复用 Stata 会话 |
| 写入 do 文件;普通文件名写到 |
| 读取 do 文件内容 |
| 追加 do 文件内容 |
| 读取 Stata text log |
| 在 Stata GUI 中安装外部包 |
| 在 Stata GUI 中显示 |
| 在 Stata GUI 中显示 |
| 读取当前数据结构、缺失摘要和样本预览并返回给 Claude |
| 查看 MCP 和 Stata 会话状态 |
do 文件和会话规则
stata_run_dofile(path)使用 do 文件绝对路径作为会话标识。同一个 do 文件路径会复用同一个 Stata 窗口。
不同 do 文件路径会创建不同 Stata 窗口。
stata_run、stata_get_results、stata_get_data_info、stata_get_data_schema默认发送到最近一次运行的 do 文件会话。如果还没有运行过 do 文件,无路径命令会提示先运行一个 do 文件。
读取当前数据结构和样本
stata_get_data_schema 会在最近 do 文件会话中生成一份 text log 快照,并把内容返回给 Claude。默认包含:
describecodebook, compactmisstable summarizelist in 1/20, abbreviate(20)noteslabel dir
可选参数:
参数 | 默认值 | 说明 |
|
| 样本预览行数,最大 1000 |
|
| 是否包含 compact codebook |
|
| 是否包含样本预览 |
|
| 是否包含缺失值摘要 |
这让 Claude 不仅能看到变量名、类型和标签,还能看到少量真实数据,从而更准确地辅助修改 do 文件。
Log 策略
stata_get_data_schema会覆盖生成自己的 schema 快照 log,因为它表示当前数据结构。其他分析结果需要 Claude 读取时,应让 Stata 写 text log,然后使用
stata_read_log读取。推荐规则:同一 Claude/MCP 会话内追加;下次新会话首次运行同一 do 文件时覆盖,避免 log 无限增长。
故障排查
pywin32 不可用
运行:
pip install pywin32COM 无法创建 Stata 实例
重新注册 Stata Automation:
Start-Process -FilePath "D:\Stata18\StataMP-64.exe" -ArgumentList "/Register" -Wait不要在 Git Bash 中直接运行带 /Register 的命令,因为 Git Bash 可能把 /Register 改写成路径。
stata_run 提示还没有 do 文件会话
先运行一个 do 文件:
stata_run_dofile(path="D:/Stata18/mcp/examples/browse_test.do")之后无路径命令会发送到这个最近会话。
Claude 不能直接看到 Stata GUI 结果
Claude 不能读取屏幕内容。需要让 Stata 写 text log,然后通过 stata_read_log 读取;或者使用 stata_get_data_schema 读取当前数据结构快照。
Related MCP server: Stata MCP Server
English
Features
Runs Stata commands in the real Stata GUI through Stata Automation COM.
Keeps one Stata session per absolute do-file path: the same do file reuses the same Stata window.
Different do files open different Stata windows, so multiple analysis tasks can stay visible.
Pathless commands are sent to the most recently used do-file session.
Supports writing, reading, appending, and running do files.
Supports reading Stata text logs so Claude can analyze command output.
Supports reading the current data structure, missing-value summary, and sample rows to help Claude edit do files.
Project layout
D:/Stata18/mcp/
├── README.md
├── pyproject.toml
├── .gitignore
├── stata_mcp.py # compatibility launcher
├── src/
│ └── stata_mcp/
│ ├── __init__.py
│ └── server.py # main MCP server
├── runtime/
│ ├── dofiles/ # default do files generated by Claude/MCP
│ └── logs/ # default logs generated or read by Claude/MCP
└── examples/ # example do filesRequirements
Windows
Stata 18 MP with Automation COM registered
Python 3.10+
Python packages:
mcp,pywin32Claude Code
Recommended install location
It is recommended to place this project inside your Stata installation directory, for example:
D:/Stata18/mcpIf Stata is installed somewhere else, place the project under that Stata directory, for example:
C:/Program Files/Stata18/mcpAdvanced users may place it in any stable directory. Wherever you install it, the Claude Code MCP configuration must point to the actual stata_mcp.py path.
Installation
Download or clone this repository into the
mcpfolder under your Stata installation directory.Install Python dependencies:
pip install mcp pywin32You may also use a virtual environment:
python -m venv .venv
.\.venv\Scripts\Activate.ps1
pip install mcp pywin32Open PowerShell as Administrator, then register Stata Automation. Adjust the path to your Stata installation:
Start-Process -FilePath "D:\Stata18\StataMP-64.exe" -ArgumentList "/Register" -WaitVerify COM:
python -c "import win32com.client; s=win32com.client.Dispatch('stata.StataOLEApp'); s.DoCommand('display 12345')"Expected result: the Stata GUI opens or is connected, and displays 12345.
Add to Claude Code (recommended)
Use claude mcp add:
claude mcp add stata -- python D:\Stata18\mcp\stata_mcp.pyIf you installed the project somewhere else, replace the path with your actual path. Example:
claude mcp add stata -- python "C:\Program Files\Stata18\mcp\stata_mcp.py"Restart Claude Code after adding the MCP server, then confirm that the stata tools are available.
Manual Claude Code MCP configuration
If you do not use claude mcp add, you can configure MCP manually with JSON:
"stata": {
"command": "python",
"args": ["D:\\Stata18\\mcp\\stata_mcp.py"]
}You can also point directly to the main server file:
"stata": {
"command": "python",
"args": ["D:\\Stata18\\mcp\\src\\stata_mcp\\server.py"]
}The compatibility launcher stata_mcp.py is usually recommended because it keeps your Claude Code configuration stable if the internal project layout changes later.
Environment variables
Most users do not need these. If your installation path or COM ProgID is unusual, you can configure:
Variable | Default | Purpose |
| auto-detected project root | MCP project directory |
| parent of MCP directory | Stata installation directory |
|
| used in README registration examples |
|
| Stata Automation ProgID |
Tools
Tool | Purpose |
| Run commands in the most recent do-file session |
| Run/reuse a Stata session by absolute do-file path |
| Write a do file; simple names go to |
| Read a do file |
| Append content to a do file |
| Read a Stata text log |
| Install external packages in the Stata GUI |
| Display |
| Display |
| Return current data structure, missing summary, and sample rows to Claude |
| Show MCP and Stata session status |
Do-file and session rules
stata_run_dofile(path)uses the absolute do-file path as the session key.The same do-file path reuses the same Stata window.
Different do-file paths create different Stata windows.
stata_run,stata_get_results,stata_get_data_info, andstata_get_data_schemaare sent to the most recent do-file session by default.If no do file has been run yet, pathless commands will ask you to run a do file first.
Reading current data structure and sample rows
stata_get_data_schema creates a text log snapshot in the most recent do-file session and returns it to Claude. By default, it includes:
describecodebook, compactmisstable summarizelist in 1/20, abbreviate(20)noteslabel dir
Options:
Option | Default | Description |
|
| Number of sample rows, max 1000 |
|
| Include compact codebook |
|
| Include sample preview |
|
| Include missing-value summary |
This lets Claude see not only variable names, types, and labels, but also a small real-data sample, which helps it write better Stata code.
Log strategy
stata_get_data_schemaoverwrites its own schema snapshot log because it represents the current data state.For other analysis output, ask Stata to write a text log, then use
stata_read_logto read it.Recommended rule: append within the same Claude/MCP session; overwrite the first time the same do file is run in a new session, so logs do not grow forever.
Troubleshooting
pywin32 is unavailable
Run:
pip install pywin32COM cannot create a Stata instance
Register Stata Automation again:
Start-Process -FilePath "D:\Stata18\StataMP-64.exe" -ArgumentList "/Register" -WaitDo not run the /Register command directly in Git Bash, because Git Bash may rewrite /Register as a path.
stata_run says there is no do-file session
Run a do file first:
stata_run_dofile(path="D:/Stata18/mcp/examples/browse_test.do")After that, pathless commands are sent to the most recent session.
Claude cannot directly see Stata GUI output
Claude cannot read screen contents directly. To let Claude analyze Stata output, write a Stata text log and read it with stata_read_log; or use stata_get_data_schema for a current data snapshot.
Available Tools
12 toolsstata_append_dofileA
向已有 do 文件追加内容
Inputs: path (required), content (required).
stata_get_data_infoB
运行 describe,并通过 session 最新运行日志将当前数据集基本信息直接返回给调用端
Inputs: no named parameters.
stata_get_data_schemaB
使用独立且每次覆盖的 schema text log,读取当前 Stata 数据集结构、缺失摘要和样本预览并返回给调用端
Inputs: sample_rows, include_sample, include_missing, include_codebook.
stata_get_resultsA
运行 return list 或 ereturn list,并通过 session 最新运行日志将 r() 或 e() 存储结果直接返回给调用端
Inputs: type (required).
stata_install_packageA
在已打开的 Stata GUI 中安装外部包,如 estout、ivreg2、rdrobust 等
Inputs: package (required), source.
stata_read_dofileA
读取 do 文件内容,返回给 Claude 检查、解释或继续修改
Inputs: path (required).
stata_read_logA
读取 Stata text log。path 留空时读取最近 session 的 last_log_path;推荐 output_format='dict',它会把日志解析成命令-结果对,便于 AI 判断报错位置。
Inputs: path, tail_lines, output_format.
stata_runA
在最近的 Stata MCP session 对应 GUI 中将 commands 作为一个完整代码块执行;每次覆盖该 session 的最新运行 text log,并自动返回实际输出。可先用 stata_session(action='set_recent') 切换目标 session。
Inputs: commands (required).
stata_run_dofileA
在 Stata GUI 中运行一个 do 文件。该工具以 session_id 表示一个复现任务/同一个 Stata GUI;同一任务的原始 do、检查 do、续跑 do 应使用同一个 session_id。每次调用都会覆盖该 session 在项目 .stata-mcp/cache/ 下的最新运行 text log,并自动返回日志内容;不会修改用户原始 do 文件。
Inputs: path (required), role, log_mode, session_id.
stata_sessionB
管理 Stata MCP session:列出、查询、销毁、切换最近 session。session_id 代表同一复现任务/同一个 Stata GUI,并关联 entry/source/current do、每个 do 的 log_paths 和 last_log_path。
Inputs: action (required), session_id.
stata_statusA
检查 Stata MCP 服务器状态、内存 session 和项目局部 .stata-mcp/cache/task_registry.json 中记录的 session/do/log 关系
Inputs: no named parameters.
stata_write_dofileA
将 Stata 代码写入 do 文件并保存到磁盘,返回文件路径;相对文件名会写入最近 session 项目的 .stata-mcp/dofiles/,不会使用 MCP runtime。
Inputs: content (required), filename.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 12 tools
Each tool targets a distinct artifact (do file, log, session, package, data metadata), and the descriptions clarify the boundaries well. A few pairs like stata_run/stata_run_dofile and stata_get_data_info/stata_get_data_schema are close enough that an agent could initially pick the wrong one.
Almost all tools follow a consistent stata_verb_noun pattern in snake_case. Minor deviations are stata_session and stata_status, which are noun-only names rather than verb_object, and stata_run which lacks an explicit object.
12 tools is well within the ideal scope for a domain-specific server. Each tool covers a meaningful part of the do-file, session, log, and data-inspection workflow without unnecessary bloat.
The tool surface covers the full write-do-file, append, read, run, read-log, install-package, and retrieve-results workflow coherently. Minor gaps exist around explicit session creation and do-file deletion, but arbitrary command execution via stata_run mitigates most dead ends.
Maintenance
Related MCP Connectors
Use your Mac, Windows or Linux computer from ChatGPT, Claude or Codex: files, commands, documents.
Connect ChatGPT, Claude, Cursor or VS Code to your Windows PC: files, terminal, browser, apps.
Remote Desktop Commander: use files, terminal and processes on your own computers from any AI app.
Control Unreal Engine to browse assets, import content, and manage levels and sequences. Automate…
Related MCP Servers
- AlicenseNot gradedqualityBmaintenanceProvides a bridge between Stata statistical software and code editors like VS Code and Cursor, enabling users to run Stata commands directly from the editor, view output in real-time, and get AI-powered assistance with Stata coding.529MIT
- AlicenseNot gradedqualityDmaintenanceConnects AI agents to a local Stata installation, enabling execution of Stata code, data inspection, graph generation, and result verification through natural language interactions.1,054 PyPI92AGPL 3.0
- AlicenseAqualityBmaintenanceEnables LLM/agent applications to execute Stata code, load and inspect data, obtain structured regression results and graphs, and manage background tasks through the Model Context Protocol.10MIT
- AlicenseAqualityBmaintenanceEnables AI agents to invoke a local Stata installation, including a Windows Stata build under Wine on Linux and macOS, for regression analysis, econometrics, paper replication, and empirical research with command guards, resource monitoring, and automatic log capture.42AGPL 3.0