Brand Voice MCP Server

brand-voice

Enforce brand writing guidelines in Claude Code.

A PostToolUse hook checks every Markdown file Claude writes or edits against your brand guidelines and signals Claude to correct violations before saving. Visual identity — colors, fonts, logo URLs — lives in the same guidelines file so Claude can reference them when generating any brand-compliant asset. Includes an MCP server for on-demand analysis and a /brand-voice-setup skill for guided onboarding.

Install

npm install -g brand-voice

Or use without installing:

npx brand-voice@latest check

Requires Node.js 18+.

Quick Start

Step 1 — Run the setup skill in Claude Code

/brand-voice-setup

The skill either reads an existing brand guidelines file you provide or asks four questions to create one. It writes brand-guidelines.md, injects a summary into CLAUDE.md, and registers the hook and MCP server in .claude/settings.json and .mcp.json.

Step 2 — Write Markdown as normal

Every time Claude uses Write, Edit, or MultiEdit on a .md or .mdx file, the PostToolUse hook runs brand-voice-check against the result. If violations are found, the hook exits with code 2 and outputs the violation list — Claude reads this feedback, corrects the violations, and retries the write automatically.

Step 3 — Iterate on your guidelines

Edit brand-guidelines.md whenever your rules change. No reinstall needed — the hook reads the file on every invocation.

How It Works

Violation types checked

• Forbidden terms — whole-word, case-insensitive match; severity: error
• Avoid terms — same matching; severity: warning
• Sentence length — configurable max words per sentence (default: 25)
• Passive voice — auxiliary + past-participle pattern detection
• Readability grade — Flesch-Kincaid grade per sentence checked against your target

Readability scores (Flesch Reading Ease, Flesch-Kincaid grade, sentence/word counts) are returned by analyze_readability.

PostToolUse hook exit codes

Exit 1 is never used by the hook (it would abort the session rather than signal a retry).

Example hook output

Brand Voice violations in quickstart.md:
  Line 12: Forbidden term: "simply"
  Line 23: Avoid term: "utilize"
  Line 31: Sentence too long: 34 words (max 25). Sentence: "Our platform..."
  Line 45: Possible passive voice: "is configured"
Fix these violations, then the file will be saved.

Brand Guidelines File

brand-guidelines.md is a plain Markdown file in your project root (or ~/.claude/ for user-scope). Keep it under 600 words / ~750 tokens so it fits cleanly in CLAUDE.md context.

Schema

# Brand Guidelines

## Persona
Who you are and who you write for.

## Tone & Voice
- Direct, honest, clear
- Person: second        ← "first" | "second" | "third"
- Voice: active         ← "active" | "passive"
- Sentences: max 25 words
- Contractions: yes     ← "yes" | "no"
- Exclamation marks: no

## Vocabulary
**Always use:** Acme, Acme Platform, APIs
**Avoid:** leverage, utilize, synergy, seamless
**Forbidden:** [competitor names, unverified performance claims]

## On-Tone Examples
> Connect your data in minutes — Acme handles the routing, you handle the insights.

## Off-Tone Examples
> Leverage our cutting-edge platform to seamlessly integrate your workflows.

## Visual Identity
- Primary color: #006EFA
- Secondary color: #0050C3
- Accent color: #FF9DFF
- Background color: #FFFFFF
- Text color: #282828
- Logo (light): https://cdn.example.com/logo-light.svg
- Logo (dark): https://cdn.example.com/logo-dark.svg
- Icon / mark: https://cdn.example.com/icon.svg
- Heading font: Inter
- Body font: Source Sans Pro

## Formatting Rules
- Heading style: sentence case
- Oxford comma: yes
- Readability target: 8th grade

## Quick Reference
Repeat your top 5 rules here. This section appears at the end of the
file where LLM attention is highest — restating key rules here counters
"Lost in the Middle" degradation in long context windows.

The ## Quick Reference section is required. It restates the most critical rules at the end of the file, where attention is strongest, to reinforce them against context-window attention drop-off.

Section name aliases: ## On-Brand Examples and ## Off-Brand Examples are accepted as aliases for ## On-Tone Examples and ## Off-Tone Examples. All other section names are case-insensitive exact matches.

See example/brand-guidelines.md for a complete working example.

Guidelines search path

The hook and CLI search for brand-guidelines.md in this order:

1. Current working directory
2. ~/.claude/brand-guidelines.md (user scope)
3. Parent directories up to the git root

Importing from an existing document

If you have a brand-guidelines.md that already follows the schema above, brand-voice import re-parses and normalizes it into the current directory:

brand-voice import /path/to/brand-guidelines.md

For a document with different section headings or prose narrative, use /brand-voice-setup instead — the skill uses Claude to translate any format into the schema.

MCP Server

The brand-voice-mcp server exposes two tools:

analyze_readability

Check a file or inline text for violations and get readability scores.

Inputs:

Returns: { filePath, passed, violations[], readabilityScores, visualIdentity }

apply_suggestions

Apply safe word-level substitutions to fix common violations (forbidden/avoid terms). Does not fix sentence length, passive voice, or readability — those require manual editing.

Inputs:

Returns: { dryRun: true, diff, changes[] } (dryRun) or { filePath, changes[], message } (live)

Tip: Call apply_suggestions with dryRun: true first to preview changes, then again without dryRun to apply them.

Distribution

CI Integration

brand-voice includes a standalone CLI for pipelines, independent of Claude Code.

Check all .md files:

npx brand-voice check

Check only files changed in the current branch:

npx brand-voice check --changed-only

GitHub Actions annotations (inline PR diff comments):

npx brand-voice check --reporter github-pr-review

Ratchet enforcement — prevent regressions without requiring a clean slate:

# Save current violation count as baseline (run once, commit the file)
npx brand-voice baseline --save

# In CI: fail only if violations increase above baseline
npx brand-voice check --baseline .brand-voice-baseline.json

Example GitHub Actions workflow:

name: Brand Voice
on:
  pull_request:
    paths: ['**.md', '**.mdx']

jobs:
  prose:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
        with: { fetch-depth: 0 }
      - run: npx brand-voice@latest check --changed-only --reporter github-pr-review

Exit code 0 = clean or within baseline. Exit code 1 = errors found or baseline exceeded.

CLI Reference

brand-voice <command> [options]

Commands:
  check [file]              Check a file or all .md files in cwd
  setup                     Print instructions to run /brand-voice-setup
  import <file>             Normalize a brand-guidelines.md into cwd
  baseline --save           Save current violation count as ratchet baseline
  vale-sync                 Check that the Vale binary is available

Check options:
  --changed-only            Only check files changed in git (requires git)
  --baseline <file>         Compare against a baseline JSON file (ratchet check)
  --reporter github-pr-review  Emit GitHub Actions annotation format

Programmatic API

import { parseGuidelines, analyzeText, loadGuidelines } from 'brand-voice';

const guidelines = loadGuidelines(process.cwd());
if (guidelines) {
  const result = analyzeText(markdownString, 'doc.md', guidelines);
  console.log(result.violations);       // Violation[]
  console.log(result.readabilityScores);
}

See src/types.ts for the full type definitions.

Requirements

• Node.js 18+ — required runtime
• brand-guidelines.md — required for enforcement; created by /brand-voice-setup
• Vale (optional) — only needed for vale-sync command; not required for the hook, MCP server, or CLI check

Contributing

1. Fork the repository and create a branch.
2. npm install && npm run build
3. npm test — runs the Jest suite under src/
4. npm run typecheck — strict TypeScript via tsc --noEmit
5. Open a pull request with a clear description of the change.

Bug reports and guideline schema improvements are welcome via GitHub Issues.

License

MIT — see LICENSE.