Styxis
Company
TRAPOT Protocol GatewayTruthound DepotData/AI Governance
TruthoundCore technology lineTruthound OrchestrationExecution and integration layerGolem / AWEHeadless Agentic Workflow EngineMabinogionProtocol resilience engine
Knowledge

Styxis Labs · Open source · Local MCP

Truthound Local — data validation MCP for CSV and Parquet

Check CSV and Parquet quality, profile columns and scan sampled PII with local MCP tools and explicit directory consent.

Truthound Local is an open-source data validation MCP server for AI agents. It validates, profiles, and scans approved local CSV and Parquet datasets without uploading raw values.

The open-source Local MCP plugin checks missing values and duplicate rows, summarizes column types and aggregate statistics, and estimates likely PII using sampling. It runs locally over stdio. Installation grants no directory access.

This open-source tool is distinct from Truthound Depot, the commercial Data/AI Governance product. It does not include Depot organizations, review, approval or managed version governance. Use the open-source tool for approved local file inspection; use the Depot product page for commercial governance adoption.

No Styxis account, API key or subscription is needed for the Local MCP server. First setup downloads dependencies; offline execution requires an already prepared runtime. Your chosen AI host can have separate requirements.

The published Local runtime is pinned to Core 3.3.1. Its stable tools are check_file, profile_file and scan_pii. Paths, column names and aggregates can still reach the host model provider. No file editing, cloud upload, arbitrary schema/rule input, or legal compliance determination is provided by these three tools.

Choose a tool for the dataset task

Every tool takes an absolute file path inside an explicitly approved directory.

TaskToolResult and boundary
Check CSV or Parquet missing values and duplicate rowscheck_fileBuilt-in check verdicts and bounded issue counts
No custom schema/rule input; inspect quality success separately from execution.
Profile a local dataset before analysis or ingestionprofile_fileInferred column types, null/unique percentages and aggregates
No sample rows; observed types do not prove schema conformance.
Screen likely PII before RAG ingestion or dataset sharingscan_piiSampled PII findings and estimated counts
Sampling can miss values; this is not masking or a complete privacy audit.

Package, MCP server and installation catalog

Truthound Python package / Core 3.3.1
The published runtime that performs validation and profiling.
Truthound Local MCP / plugin 0.1.7
Three local file tools exposed to an MCP host, using Core 3.3.1.
Styxis plugin marketplace
The installation catalog pointing to the public Local plugin source.

Verified with the published package

A three-row, two-column synthetic dataset was checked as CSV and Parquet. Six calls verified failed quality checks for nulls/duplicates, aggregate profiles and sampled email findings. An outside-directory call was denied and raw values were absent from results.

Reproduce the input, commands and resultsRead the actual JSON results

Execution evidence for Core 3.3.1; not a PII accuracy benchmark or an AI search visibility score.

Install in Claude Code

claude plugin marketplace add seadonggyun4/styxis-plugin
claude plugin install truthound@styxis-plugin

After installation, approve the exact data directory through the setup skill.

Sources, installation and evidence

Evidence date:

Explore commercial Truthound Depot