Truthound
Local data quality checks for CSV and Parquet files, run from an AI agent
- Pricing
- Open source
- Platforms
- CLI, self-hosted, macOS
- Runs
- Locally
- Version
- 3.3.4
- Listing updated
- 3 days ago
About Truthound
Truthound is an open-source MCP server for analytics and data. It runs locally on your machine and is started by your AI client.
This server wraps the Truthound data-validation library so an agent can inspect a data file on your own machine. It runs over stdio, takes a path inside a folder you approve, and returns summaries rather than raw rows. URLs, glob patterns, parent-directory jumps and symlinks under an allowed folder are refused.
The default Local profile has three tools: check_file for missing values and duplicate rows, profile_file for row and column counts, type families and null or unique ratios, and scan_pii, which samples the first 1,000 non-null values per column so reported counts are estimates. A Core profile adds schema learning, drift comparison, masking and validation with custom rules, each behind separate permission approvals. The docs note that a failed quality check still returns ok=true, and that full MCP parity with the Core library is not yet verified.
Supported formats are CSV, Parquet, JSON and line-delimited JSON. Installation goes through the Styxis plugin marketplace for Codex CLI and Claude Code and needs Git, uv and Python 3.11 to 3.13.
Install
Run in a terminal
claude mcp add truthound -- uvx truthoundAdd to claude_desktop_config.json
{
"mcpServers": {
"truthound": {
"command": "uvx",
"args": [
"truthound"
]
}
}
}Add to ~/.cursor/mcp.json
{
"mcpServers": {
"truthound": {
"command": "uvx",
"args": [
"truthound"
]
}
}
}Add to .vscode/mcp.json
{
"servers": {
"truthound": {
"type": "stdio",
"command": "uvx",
"args": [
"truthound"
]
}
}
}Add to ~/.codeium/windsurf/mcp_config.json
{
"mcpServers": {
"truthound": {
"command": "uvx",
"args": [
"truthound"
]
}
}
}Add to your client's MCP settings (most use this shape)
{
"mcpServers": {
"truthound": {
"command": "uvx",
"args": [
"truthound"
]
}
}
}Tools this server provides (9)
Taken from the project documentation; the running server may differ.
- check_file
- Checks a file for missing values and duplicate rows.
- profile_file
- Summarises rows, columns, type families and null or unique ratios.
- scan_pii
- Samples columns for likely personal-data types.
- validation_check
- Core: validates a file against specified rules.
- data_profile
- Core: summarises column types and quality.
- privacy_scan
- Core: finds columns suspected of holding PII.
- data_mask
- Core: masks values such as emails by hashing without altering the source file.
- schema_learn
- Core: learns a schema from data.
- drift_compare
- Core: compares a baseline and a new batch for drift.
What it does
- Missing and duplicate checks
- check_file runs the built-in null and duplicate checks and reports pass or fail with an issue count.
- Column profiling
- profile_file returns row and column counts, type families and null or unique ratios.
- Sampled PII scan
- scan_pii flags likely personal-data columns using the first 1,000 non-null values per column.
- Folder-scoped access
- Only folders you approve are readable, and installing the plugin grants no data access by itself.
- Core profile
- Adds schema learning, drift comparison, masking and rule-based validation with extra permissions.
- Structured errors
- Stable error codes such as access_denied and file_too_large, without internal exception text.
Good for
- Check a CSV export for nulls and duplicate rows before loading it
- Profile an unfamiliar Parquet file's columns and types
- Find columns that look like emails or other personal data
- Compare a new batch against a reference dataset for drift (Core)
Pricing
Open source
Apache-2.0 library; runs locally
Prices change. Check truthound-docs.tech before you buy.
Truthound FAQ
Does it send my data anywhere?
No. The server runs locally over stdio and calls the library on your computer; the docs say it is not a remote HTTP or OAuth connector.
Which file types work?
CSV, Parquet, JSON, NDJSON and JSONL.
How is PII counted?
By sampling the first 1,000 non-null values per column, so the numbers are estimates.
What do I need installed?
Git, uv and Python 3.11 to 3.13, plus a client that supports plugins such as Codex CLI or Claude Code.
Is Truthound free?
Yes. Truthound is open source, so you can use and self-host it for free.
How do I add Truthound to Claude Code?
Run this command in your terminal: claude mcp add truthound -- uvx truthound
What are the best alternatives to Truthound?
The closest alternatives to Truthound are REA, Codebase Memory, GitHub, Serena MCP: the IDE for your agent and XERJ.
Alternatives to Truthound

REA
MCP serverCLI and MCP server that lets coding agents reverse engineer apps and binaries
Runs locally- Open source
- 464K GitHub stars
- Developer tools

Codebase Memory
MCP serverLocal MCP server that builds a code knowledge graph for AI coding agents
Runs locally- Open source
- 48.7K GitHub stars
- Knowledge and memory

GitHub
MCP serverOfficial MCP server connecting AI assistants to GitHub repos, issues and workflows
Not checked yet- Remote
- Open source
- 34.5K GitHub stars
- Developer tools

Serena MCP: the IDE for your agent
MCP serverMCP toolkit giving coding agents symbol-level code retrieval and editing
Runs locally- Open source
- 31.5K GitHub stars
- Developer tools

XERJ
MCP serverLocal search engine for AI agents with cited answers over your code and docs
Runs locally- Open source
- 19.2K GitHub stars
- Knowledge and memory

Agent-Native Analytics
MCP serverOpen-source analytics app where an AI agent builds charts and dashboards from your data
- Remote
- Open source
- 7.7K GitHub stars
- Analytics and data