Use Bloomberry's AI Writing Scan MCP server in Claude or Cursor

Bloomberry is bringing MCP-native AI-writing guardrails to employee advocacy. This guide covers local setup for Claude Desktop and Cursor.

What the MCP server does

The Bloomberry AI Writing Scan MCP server checks text against Bloomberry's production AI-writing signal corpus and returns structured findings: risk scores, specific flagged signals, and rewrite suggestions. It is a writing quality tool, not an authorship detector.

Corpus

The V1 scanner uses the production enforcement corpus:

• 5,698 production AI-writing signal entries enforced at runtime
• The Bloomberry AI Sentence DNA research corpus contains 7,622 total audited entries (public label: 7,400+ catalogued AI-writing signal entries). The research corpus includes additional entries not yet in production enforcement.

Available tools

scan_ai_sounding_text

Full scan. Returns risk score, risk label, findings, banned phrase matches, sentence-DNA matches, rewrite suggestions, and corpus stats.

// Input
{
  "text": "string",
  "strictness": "low | medium | high",   // optional, default: medium
  "contentType": "linkedin_post | blog | email | executive_post | general"  // optional
}

// Output
{
  "riskScore": 0-100,
  "riskLabel": "low | medium | high",
  "summary": "string",
  "findings": [...],
  "bannedPhraseMatches": [...],
  "sentenceDnaMatches": [...],
  "suggestedFixes": [...],
  "phraseLibraryMode": "production-enforcement-corpus",
  "productionSignalEntryCount": 5698,
  "researchAuditedSignalEntryCount": 7622,
  "disclaimer": "This does not prove whether AI wrote the text. It only identifies language patterns that may sound generic, synthetic, or off-brand."
}

check_banned_phrases

Checks text against the production phrase corpus. Returns matches with severity and replacement guidance.

// Input
{ "text": "string" }

// Output
{
  "matches": [...],
  "phraseLibraryMode": "production-enforcement-corpus",
  "productionSignalEntryCount": 5698,
  "disclaimer": "..."
}

analyze_sentence_dna

Detects sentence structures that make writing sound synthetic, generic, or overly polished.

// Input
{ "text": "string" }

// Output
{
  "sentenceDnaMatches": [...],
  "riskScore": 0-100,
  "riskLabel": "low | medium | high",
  "disclaimer": "..."
}

Installation and local run

The MCP server ships with the Bloomberry backend. No separate npm install is required if you have the repo cloned.

"color:#6e7681"># From the repo root
"color:#79c0ff">node backend/mcp/bloomberryAiWritingScanServer.js

Or use the npm script:

"color:#79c0ff">npm run mcp:ai-writing-scan

Connect in Claude Desktop

Add the following to your claude_desktop_config.json (usually at ~/Library/Application Support/Claude/ on macOS):

{
  "mcpServers": {
    "bloomberry-ai-writing-scan": {
      "command": "node",
      "args": ["/absolute/path/to/Xreply/backend/mcp/bloomberryAiWritingScanServer.js"]
    }
  }
}

Replace /absolute/path/to/Xreply with the actual path to your cloned repo. Restart Claude Desktop after saving.

Connect in Cursor

Add to .cursor/mcp.json in your project root (or the global Cursor MCP config):

{
  "mcpServers": {
    "bloomberry-ai-writing-scan": {
      "command": "node",
      "args": ["./backend/mcp/bloomberryAiWritingScanServer.js"]
    }
  }
}

Example prompts

Once connected, you can call the tools from Claude or Cursor with natural language prompts:

"Scan this LinkedIn post for AI-sounding language."

"Check this draft for banned phrases and suggest alternatives."

"Analyze the sentence DNA of this post — does it follow AI cadence patterns?"

"Give me a full AI writing scan of this executive thought leadership post."

Security and data scope

• Local and stateless. V1 runs locally via stdio. No data leaves your machine to a Bloomberry server.
• No authenticated data. The MCP server does not connect to Supabase, Voice Memory, workspace data, or any authenticated Bloomberry API.
• No Voice Memory comparison. Comparing against an individual's voice profile requires authentication. That feature is not in V1.
• No write actions. The server only reads and scans — no writes, no posts, no scheduling.

Troubleshooting

Server does not start

• Confirm Node.js 18+ is installed: node --version
• Run npm install inside backend/ to ensure MCP SDK dependencies are present
• Check the path in your config is absolute and correct

Tools do not appear in Claude or Cursor

• Restart Claude Desktop or Cursor after updating the config file
• Verify the server starts cleanly by running it manually first
• Server startup logs go to stderr — they will not appear in MCP stdout

stdout pollution / JSON-RPC errors

• The server uses a silentIsAIDialect() wrapper to suppress console.log output from aiDialect.js. If you see unexpected stdout, check that no other imports are writing to stdout.

FAQ