MCP SERVERLet an agent run a Stata do-file on your own machine, read the log it produced, and look up the syntax of the command that failed.
LABSCO SUMMARY
Execution is subprocess-based rather than session-based, so each run is a do-file in and a log out — reproducible, and easy to inspect after the fact. The guard and the RAM ceiling exist because the caller is a model: shell escapes and file deletion are refused before Stata ever sees them.
An MCP server that hands a licensed local Stata install to an agent. The agent writes a do-file, the server runs it through the Stata CLI as a subprocess and returns the execution log; separate tools describe a data file before it is used and read back .log or .smcl output afterwards. A command guard sits in front of execution, and the same tools are also exposed as a standalone CLI toolkit.
stata_do.dta, .csv, .tsv, .psv, .xlsx and .xls — get_data_infohelpread_log!, shell and erase before they reach StataSTATA_MCP__ environment variables overriding individual settingsYour own Stata licence, installed at the default path, with the Stata CLI present on macOS and Linux. uv provides the runtime: uvx stata-mcp install --all writes the config for every supported client, or uvx stata-mcp install -c cc for one of them; the published package is stata-mcp 1.22.0 on PyPI and runs over stdio. uvx stata-mcp doctor reports whether the OS, Python, dependencies, Stata CLI and execution path all check out before you debug a client. Claude Science runs servers in a sandbox that blocks the home directory, so it needs the uv tool paths allowlisted in its own config. Licensed AGPL-3.0.
One command — uvx stata-mcp
USE CASES
Point the agent at the dataset, have it inspect the variables before writing anything, then run the do-file and read the log to compare coefficients against the published table.
Run it, read the error out of the log, look up the offending command's documented syntax from your own Stata install, and try again — without leaving the conversation to open the manual.
Get descriptive statistics and a preview for the file — including .csv and .xlsx sources, not only .dta — before committing to a specification.
Ask for a method, have the agent write and run the do-file, then read the output back with the command's own documentation alongside it.
Counted once with only startup setup. Other versions, modes or optional credentials may expose a different list.
get_data_infoGet descriptive statistics and a data preview for a supported data file (.dta, .csv, .tsv, .psv, .xlsx, .xls, .sav, .zsav). Returns overview, variable details, and optional head rows filtered by requested variables. Use when you need to understand a dataset or have no prior knowledge of the data.
helpRetrieve documentation and usage information for a Stata command. Use when you need to understand a command's syntax, options, or troubleshoot errors before running it.
read_logRead a Stata log file (.log or .smcl) and return its content. Supports full, core, and dict output formats. Use lines to return only the first/last N lines.
stata_doExecute a Stata do-file and return the execution log. Accepts a do-file path, runs it via the configured Stata executable, and can optionally read log content only when return-code errors are detected. Use when you need to run Stata commands, perform regression or statistical analysis, or execute a do-file.
BEFORE YOU INSTALL
4 tools when we connected. A fifth tool, ado_package_install, registers only with stata-mcp server --unsafe. On Windows, help is excluded and get_data_info is hidden by default, leaving 2.
uvx on PATH
CONTEXT COST
Rated by the size of the 4 tool definitions that ride along on every turn, not by how many there are
INSTALLATION
uvx stata-mcp{
"mcpServers": {
"stata-mcp": {
"command": "uvx",
"args": [
"stata-mcp"
],
"env": {
"STATA_MCP__CWD": "<YOUR_STATA_MCP_CWD>",
"STATA_CLI": "/Applications/Stata/StataMP"
}
}
}
}{
"mcpServers": {
"stata-mcp": {
"command": "uvx",
"args": [
"stata-mcp"
],
"env": {
"STATA_MCP__CWD": "<YOUR_STATA_MCP_CWD>"
}
}
}
}{
"mcpServers": {
"stata-mcp": {
"command": "uvx",
"args": [
"--directory",
"<YOUR_STATA_MCP_CWD>",
"stata-mcp"
],
"env": {
"STATA_MCP__CWD": "<YOUR_STATA_MCP_CWD>"
}
}
}
}{
"mcpServers": {
"stata-mcp": {
"command": "uvx",
"args": [
"stata-mcp"
]
}
}
}[mcp_servers.stata-mcp]
command = "uvx"
args = ["stata-mcp"]
env = { STATA_MCP__CWD = "<YOUR_STATA_MCP_CWD>" }{
"mcp": {
"stata-mcp": {
"type": "local",
"command": [
"uvx",
"stata-mcp"
],
"env": {
"STATA_MCP__CWD": "<YOUR_STATA_MCP_CWD>"
}
}
}
}{
"mcp": {
"servers": {
"stata-mcp": {
"command": "uvx",
"args": [
"stata-mcp"
],
"env": {
"STATA_MCP__CWD": "<YOUR_STATA_MCP_CWD>"
}
}
}
}
}mcp_servers:
stata-mcp:
command: "uvx"
args: ["stata-mcp"]
env:
STATA_MCP__CWD: "<YOUR_STATA_MCP_CWD>"uvx stata-mcp install -c piMCP servers
Tsuchijo
Archerkattri
ProgramComputer
benedictdebrah