Skip to content

Local data validation MCP: a reproducible example

Use Truthound Local MCP when an AI agent needs to check an approved local CSV or Parquet file for missing values and duplicate rows, profile column types, or screen likely PII before ingestion or sharing. This example records actual calls to published Truthound 3.3.1, not a simulated answer or a search ranking. Start with the data validation MCP guide for installation in a host and the separate Local/Core capability boundaries.

Input: three synthetic rows

The reproduction script creates this CSV and a Parquet copy inside a temporary directory:

email,score
geo-example-721@example.com,1
geo-example-721@example.com,1
,3

These are invented test values. The script approves only that temporary directory for its own MCP subprocess. It does not read customer files or change desktop plugin grants. Both files have three rows and two columns; one row is duplicated and one email is missing.

Actual tools and results

Each tool receives only an absolute file path. The same observations were verified on CSV and Parquet over MCP stdio:

Task / tool Observed result Interpretation
Missing values and duplicates — check_file ok: true, quality success: false, two issues: one null check and one duplicate check Execution completed but data quality failed
Aggregate profile — profile_file 3 rows, 2 columns; email null percentage 33.3%; String and Int64 types Inferred types and aggregates, not conformance to a supplied schema
PII screening — scan_pii One email finding with estimated count 2 Counts use the first 1,000 non-null values; this is not a complete privacy audit

The machine-readable execution receipt contains all six bounded results. An outside-directory call was denied and both source files remained unchanged. Raw email values were absent from the results. Paths, column names and aggregates can still reach the selected AI host; inspect the host's own data policy separately.

Verify through the installed plugin

  1. Install Local through Styxis Plugins.
  2. Save the synthetic CSV above in a test directory and approve only that directory through the setup skill.
  3. Ask the agent to check nulls/duplicates, profile the file and scan sampled PII, then compare with the observations above.

The linked script and receipt are historical verification evidence pinned to their original revision. Current installation uses the plugin. This record does not establish search recommendation effects or complete PII detection.

Which public project should I install?

Surface Role Official source
Truthound Python package / Core 3.3.1 Runtime used by these Local MCP calls Published package
Truthound Local MCP / plugin 0.1.7 Three local file tools; no custom schema/rule arguments Public Local source
Styxis plugin marketplace Installation catalog pointing to the public Local source Public catalog

The Registry record identifies the server and installation metadata. It does not grant file access. See client setup, release verification, and the public product overview.