01

One analysis engine, five ways to use it

WEB

Web for people

Paste text, inspect bounded tables, compare results, and export evidence in a visual interface. Large tables are marked when only the returned page is shown.

HTTP

API for apps and agents

Call eight stateless JSON operations with OpenAPI discovery, stable errors, and no API key.

CLI

CLI for automation

Analyze files, URLs, inline text, or stdin locally and return tables, JSON, or CSV.

MCP

MCP for AI agents

Expose the same eight local read-only operations as discoverable tools over stdio.

WEBMCP

Claim-aware version review

ChatGPT uses five page tools to map claim drift and submit a complete 3–5 item review plan. Every exact diff remains under human approval. Open the workspace →

02

Connect the npm package to an MCP client

The MCP command starts a local stdio server. Text supplied to a tool stays inside that process; only explicit public URL inputs use the network.

npx --yes textanalysis-tools@0.2.0 mcp
{
  "mcpServers": {
    "textanalysis": {
      "command": "npx",
      "args": ["--yes", "textanalysis-tools@0.2.0", "mcp"]
    }
  }
}

03

Eight focused operations

Each tool has a bounded input schema and structured output. Agents can choose a focused operation instead of parsing a general-purpose report.

analyze_text

read-only · structured JSON · local execution

word_frequency

read-only · structured JSON · local execution

keyword_density

read-only · structured JSON · local execution

ngram_analysis

read-only · structured JSON · local execution

bag_of_words

read-only · structured JSON · local execution

compare_texts

read-only · structured JSON · local execution

tfidf

read-only · structured JSON · local execution

text_similarity

read-only · structured JSON · local execution

04

Three practical agent recipes

Use these sequences as auditable building blocks. Keep source identifiers beside every result and make decisions from measured fields, not generated impressions.

keyword_density → ngram_analysis → analyze_text

Repetition and density audit

Find repeated tracked phrases, recurring wording, and terms that sit well above the fitted frequency model.

  1. Call keyword_density with the phrases the draft is expected to cover.
  2. Call ngram_analysis with ngramSize 2 or 3 to surface unplanned repetition.
  3. Use analyze_text to inspect above-model terms, then cite counts and rates instead of inventing a quality score.
compare_texts

Draft versus baseline regression

Detect vocabulary, length, and recurring-phrase changes between an approved baseline and a new draft.

  1. Send the approved document as a and the new draft as b with the same language and stop-word settings.
  2. Rank wordChanges and bigramChanges by absolute shareDelta, not raw count alone.
  3. Report the changed metric, both source values, and the normalized delta; route intentional editorial changes for human approval.
text_similarity → compare_texts

Duplicate and near-duplicate detection

Screen a candidate against an existing document, then explain why a pair looks similar.

  1. Call text_similarity with method tfidf and a project-calibrated threshold; do not treat one universal cutoff as ground truth.
  2. For flagged pairs, call compare_texts to expose shared and changed vocabulary rather than relying on the score alone.
  3. Persist the cosine score, method, input identifiers, and top contributing terms, then send borderline cases to human review.

05

Use OpenAPI when HTTP is the better transport

The public API exposes the same analysis primitives for remote applications. Use the machine-readable contract for tool generation and the concise capability map for discovery.

Machine-readable discovery

https://textanalysis.tools/openapi.json
https://textanalysis.tools/llms.txt

06

Deterministic analysis, not generated judgment

  • No LLM is used to calculate the results.
  • Counts and formulas remain inspectable.
  • Local CLI and MCP text is not sent to textanalysis.tools.
  • Public URL inputs require a network request to that page.
  • Results are diagnostics, not guarantees about ranking or content quality.

WEB · API · CLI · MCP

Choose the interface that matches the workflow

Use MCP for local agent tools, CLI for shell automation, API for remote applications, or the web interface for manual review.

Read CLI and MCP documentation →