# Vocabulary Forge MCP

> Vocabulary Forge builds a complete human-authentic writing profile for your agent. Instead of asking an AI model to 'write like a person,' this MCP forces it through five decision pivots: defining a specific voice, mapping tone shifts, banning common AI words, integrating colloquialisms, and committing structural signatures. It gives you the vocabulary blueprint that defeats detection.

## Overview
- **Category:** productivity
- **Price:** Free
- **Tags:** vocabulary, humanization, ai-detection, anti-detection, voice-design, structured-reasoning, decision-pivots, multilingual

## Description

AI-generated text always carries invisible tells; they're usually not in grammar. The signal is in the vocabulary itself—the tendency for a model to reach for the same overly formal words repeatedly. This MCP lets your agent build an entire voice profile before it writes anything. You define who you are: age, habits, and specific pet peeves. Then, you map how that voice shifts tone across different sections of text, like moving from a casual opening thought to a technical body paragraph. Crucially, the tool forces you to ban known AI signal words—words like 'leverage' or 'comprehensive.' It also adds native colloquialisms and structural signatures so the output sounds genuinely irregular and human. Because Vinkius manages this process inside an isolated sandbox, you get full visibility into exactly how your agent is building that profile using Vinkius AI Analytics; nothing happens in the dark. This deep control over vocabulary makes it reliable for any language.

## Tools

### forge_vocabulary
Builds a structured, multi-layered human profile by anchoring a specific voice, mapping register shifts, purging AI signal words, integrating colloquialisms, and committing signature expressions.

## Prompt Examples

**Prompt:** 
```
I need to write a blog post about AI tools. Build me a vocabulary profile that sounds professional and engaging.
```

**Response:** 
```
Verdict: VOICE_ABSENT. "Professional and engaging" is a mood, not a voice. Who is writing this blog post? A former developer turned product manager who writes like she's explaining something to a smart friend over coffee? A cynical journalist who's seen too many AI hype cycles? An enthusiastic CTO who writes in short bursts and uses too many em-dashes? Define the person first.
```

**Prompt:** 
```
Voice: cynical tech journalist, mid-30s, hates buzzwords, short punchy sentences. Banned: delve, leverage, comprehensive, robust, furthermore, tapestry, landscape, seamless, utilize, harness, holistic, paradigm. Colloquialisms: look, honestly, here's the thing, don't, can't. Signatures: single-word paragraph closers, question openers, dashes over semicolons. Register: provocative opener, skeptical body, imperative close.
```

**Response:** 
```
Verdict: VOICE_AUTHENTIC. Voice anchored — cynical tech journalist with specific habits (short sentences, periods, dashes). Register mapped — 3 distinct zones (provocative → skeptical → imperative). Anti-vocabulary: 12 banned words including 8 known AI signals. Colloquialisms: 5 discourse markers native to English. Signatures: 3 structural tics (single-word closers, question openers, dash preference). This vocabulary profile defeats detection. Write with it.
```

**Prompt:** 
```
Voice: senior product manager who communicates clearly. Register: professional throughout. Banned: delve, leverage. Colloquialisms: none — this is formal writing. Signatures: none needed.
```

**Response:** 
```
Multiple rejections. (1) VOICE_ABSENT — "senior PM who communicates clearly" is a LinkedIn headline, not a voice. (2) FLAT_REGISTER — "professional throughout" is the strongest AI signal. Map 3 zones. (3) AI_LEXICON — only 2 banned words, minimum is 8. (4) STERILE — "none" for colloquialisms = no human roughness. (5) FORGETTABLE — "none needed" for signatures = no identity. Fix all five.
```

## Capabilities

### Define a specific human identity
Creates a detailed persona (age, habits, pet peeves) that anchors the writing to an actual person.

### Map tonal shifts for consistency
Defines mandatory tone changes across three or more distinct sections of text, preventing flat, monotonous prose.

### Purge AI-signal vocabulary
Forces the agent to exclude a list of banned words known for being overused by large language models.

### Integrate native colloquialisms
Adds natural contractions, idioms, and discourse markers specific to the target language.

### Commit structural signatures
Inserts recurring phrases or structural tics that act as a unique fingerprint for the voice.

## Use Cases

### Revising stale corporate reports
An internal communications team needs to update 50 quarterly reports. Instead of letting the AI write them in the standard 'Furthermore, the organization has demonstrated robust growth' style, they use this MCP to ban those specific buzzwords and inject a more conversational, human voice.

### Drafting technical deep dives
A developer needs to explain complex system architecture. They feed the tool their desired 'voice'—a cynical but expert engineer. The resulting text maintains a sharp, skeptical tone and uses specific, non-AI jargon throughout.

### Creating localized marketing campaigns
A global brand is launching in Brazil. They use the MCP to ensure the copy integrates colloquialisms native to Brazilian Portuguese, bypassing generic language patterns that would flag it as machine-generated content.

## Benefits

- Stops flat text. By mapping your tone shifts, you ensure the writing changes energy—from casual intro to technical body—so it doesn't sound like a press release.
- Defeats signal words. You ban specific AI lexicon (like 'leverage') and replace them with domain-specific alternatives before the agent writes a single word.
- Authentic identity. Instead of general roles, you define a person, giving the writing actual habits, preferences, and structural quirks that make it feel unique.
- Global readiness. The tool doesn't rely on English rules; its pivots apply universally to languages like French or Portuguese, making your content strategy truly multilingual.
- Better consistency. By committing signature expressions—recurring phrases—you give the output a recognizable pattern that fingerprints the voice, even without a byline.

## How It Works

The bottom line is your agent gets a full style guide built right into its operating parameters before generating any text.

1. First, you must define five distinct parameters: the specific human persona, the three tone zones, at least eight banned words, native colloquialisms, and signature expressions.
2. The tool runs a validation check against your profile. It confirms that all criteria are met—for instance, checking that the voice is a person, not just a role.
3. If successful, it generates an authenticated vocabulary blueprint that guides the agent to write content that passes human-detection standards.

## Frequently Asked Questions

**Does Vocabulary Forge write content?**
No. Vocabulary Forge generates zero content. It forces the AI agent to build a complete vocabulary PROFILE — voice, register, banned words, colloquialisms, signatures — before writing anything. The profile then guides the agent's word choices. The tool validates the profile's depth and consistency, not the final text.

**What is the anti-vocabulary and why is it the most important part?**
The anti-vocabulary is the list of words the voice MUST NEVER use. AI detectors work by scanning for the presence of ~30-40 signal words that appear in AI output 10-100x more than in human text. If those words are absent, the detector's primary signal disappears. Banning 'delve', 'leverage', 'comprehensive', 'robust', 'furthermore' does more to defeat detection than adding human words. Absence is stronger than camouflage.

**Does it work for languages other than English?**
Yes — any language. The five pivots are universal: every language has human voices, register shifts, AI-overused words, colloquialisms, and signature patterns. The engine validates depth and consistency without prescribing English-specific rules. For Portuguese, ban 'além disso', 'abrangente', 'robusto'. For French, ban 'en outre', 'exhaustif'. For Japanese, identify the keigo/casual register shifts your voice uses. The colloquialisms must be native to the target language.

**How does Vocabulary Forge secure the unique voice profile I create?**
Your data never sits on a disk because credentials pass through our zero-trust proxy. Every single call is protected by Vinkius, which executes the tool inside an isolated sandbox and generates a cryptographically signed audit trail for accountability.

**Can Vocabulary Forge integrate with multiple AI clients like Cursor or Claude?**
Yes, you connect once using your preferred MCP-compatible client. That single connection lets you use this MCP across the entire catalog of tools, meaning it works regardless of which agent you're running.

**Will using Vocabulary Forge significantly increase my token usage?**
No, Vinkius handles native token optimization for every call. This built-in feature cuts your overall token consumption by up to 60% compared to standard tool runs.

**What should I do if the `forge_vocabulary` tool rejects my profile?**
It means your profile is still carrying an AI signal. You must review and fix the specific pivot that failed—usually by making your banned word list longer or defining a more specific, non-role-based voice.

**Is Vocabulary Forge only useful for long-form articles?**
No, this MCP builds an authentic profile for any language and content length. It forces the agent to maintain consistency across all text, whether it’s a quick memo or a full report.