Verto Intel MCP Server

MCP (Model Context Protocol) server that connects Claude Desktop to Verto Intel Bot - your competitive intelligence and content generation system with 20+ specialized tools.

Quick Start

1. Add to Claude Desktop Config

Open your Claude Desktop configuration file:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

Add the following:

{
  "mcpServers": {
    "verto-intel": {
      "command": "npx",
      "args": ["-y", "verto-intel-mcp"],
      "env": {
        "VERTO_API_URL": "https://your-verto-bot.up.railway.app",
        "VERTO_ADMIN_PASSWORD": "your-admin-password"
      }
    }
  }
}

2. Restart Claude Desktop

After saving the config, restart Claude Desktop to load the MCP server.

Features Overview

This MCP server provides 21+ tools organized in 5 categories:

• 🤖 9 Agent Execution Tools - Direct access to all specialized AI agents
• 📝 3 Prompt Management Tools - Version and customize agent prompts
• ✍️ 6 Content Generation Tools - High-level content creation workflows
• 📊 3 Scan & Source Management Tools - Monitor intelligence sources
• 📁 5 Resources - Read-only data access

🤖 Agent Execution Tools

Direct access to all 9 specialized agents:

list_agents

View all available agents with metadata (name, description, prompt version, custom status).

Example:

Use list_agents to see what agents are available

execute_web_scout

Research topics using Exa Search.

Parameters:

• topic (required) - Research topic
• maxSources (optional) - Max sources to return
• recencyDays (optional) - How recent sources should be

Returns:

• 5-10 relevant sources with URLs
• Key facts and statistics
• Content angles for CMOs

Example:

Use execute_web_scout to research "AI pricing strategies for SaaS 2024"

execute_doc_parser

Parse and extract from documents.

Parameters:

• url (optional) - URL to parse
• text (optional) - Raw text to parse

Returns:

• Key arguments
• Data points with evidence
• Metadata (title, author, date)

Example:

Use execute_doc_parser with url "https://example.com/article"

execute_scorer

Score content for B2B marketing value (0-100).

Parameters:

• title (required) - Content title
• content (required) - Content body
• url (optional) - Source URL

Returns:

• Total score (0-100)
• Badge (🏆 GOLD, 🥈 SILVER, 🥉 BRONZE)
• Classification (Must-Read, Worth Reading, Skip)
• Key insight for CMOs

Example:

Use execute_scorer to rate this article

execute_linkedin_writer

Generate LinkedIn posts in Sloane Bishop voice.

Parameters:

• topic (optional) - Topic to write about
• content (optional) - Source content
• focusAngle (optional) - Specific angle

Returns:

• 150-200 word LinkedIn post
• Sloane Bishop voice (outcome-obsessed, math-backed, no buzzwords)
• Hook, tension, insight, proof, CTA, hashtags

Example:

Use execute_linkedin_writer to create a post about efficient growth

execute_article_writer

Generate thought leadership articles.

Parameters:

• topic (optional) - Topic
• content (optional) - Source content
• focusAngle (optional) - Specific angle

Returns:

• 600-800 word thought leadership article
• Executive-focused, data-driven
• Structured with introduction, insights, conclusion

Example:

Use execute_article_writer to write about modern marketing tech stacks

execute_voice_critic

Critique content for voice consistency.

Parameters:

• content (required) - Content to critique
• contentType (optional) - linkedin_post or article

Returns:

• Voice consistency score (0-100)
• Specific violations
• Improvement suggestions

Example:

Use execute_voice_critic to review this LinkedIn post

execute_fact_checker

Verify factual accuracy with source citations.

Parameters:

• content (required) - Content to check
• sourceUrls (optional) - Known source URLs

Returns:

• List of factual claims
• Verification status (Verified, Likely True, Unverified, Questionable)
• Source citations

Example:

Use execute_fact_checker to verify the claims in this article

find_trends

Identify emerging, growing, peak, and declining trends across multiple sources.

Parameters:

• sources (required) - Array of content/articles to analyze
• timeframe (optional) - Time window in days (default: 30)

Returns:

• Trends with signal strength (emerging/growing/peak/declining)
• Evidence from sources
• CMO implications
• Contrarian interpretations
• Meta-narrative synthesis

Uses:

• Frequency analysis - How often mentioned
• Velocity analysis - Rate of increase/decrease
• Authority analysis - Source credibility
• Sentiment analysis - Positive/negative momentum

Example:

Use find_trends with sources array from recent web_scout results

📝 Agent Prompt Management Tools

Version and customize agent prompts:

get_agent_prompt

View current prompt and version history.

Parameters:

• agentName (required) - Agent name (e.g., linkedin_writer)

Returns:

• Current active prompt
• Version history with changelogs
• Creation dates and authors

Example:

Use get_agent_prompt for agentName "linkedin_writer"

update_agent_prompt

Create new prompt version (versioned with changelog).

Parameters:

• agentName (required) - Agent name
• newPrompt (required) - New prompt text
• changelog (optional) - Description of changes

Returns:

• New version number
• Active prompt text
• Changelog entry

Example:

Use update_agent_prompt to make linkedin_writer more aggressive

test_agent_prompt

Test prompts safely without activation.

Parameters:

• agentName (required) - Agent name
• testPrompt (required) - Prompt to test
• testInput (required) - Test input data

Returns:

• Test execution result
• Note that prompt was NOT activated
• Execution metadata

Example:

Use test_agent_prompt to try a new prompt for web_scout

✍️ Content Generation Tools

High-level content creation workflows:

create_content_from_url

Generate content from article URL.

Parameters:

• url (required) - Article URL
• outputType (optional) - linkedin_post, article, or both (default)
• tone (optional) - default, aggressive, educational, storytelling

Example:

Use create_content_from_url with url "https://techcrunch.com/article" and outputType "both"

create_content_from_topic

Research topic → generate content.

Parameters:

• topic (required) - Topic to research
• outputType (optional) - Content type
• tone (optional) - Tone

Example:

Use create_content_from_topic with topic "Modern data stack for marketing teams"

create_content_from_text

Generate from provided text.

Parameters:

• text (required) - Source text
• title (optional) - Content title
• outputType (optional) - Content type
• tone (optional) - Tone

Example:

Use create_content_from_text with the text I just provided

create_content_from_file

Generate from local file (TXT, MD, CSV).

Parameters:

• filePath (required) - Absolute file path
• outputType (optional) - Content type
• tone (optional) - Tone

Note: For PDF/DOCX, attach file to Claude chat instead.

create_content_multi_agent

Advanced multi-agent workflow with scoring & voice check.

Parameters:

• inputType (required) - url, file, or topic
• sourceValue (required) - URL, path, or topic string
• outputType (optional) - linkedin, article, or both

Orchestrates:

1. web_scout (if topic) or doc_parser (if URL/file) - Gather content
2. scorer - Score content value (0-100)
3. linkedin_writer + article_writer - Generate both formats
4. voice_critic - Validate voice consistency

Returns:

• LinkedIn post + article
• Voice check results
• Scoring data
• Workflow metadata (time, tokens, steps)

Example:

Use create_content_multi_agent with inputType "topic" and sourceValue "AI pricing strategies"

Legacy Tools

• generate_linkedin_post - Simple LinkedIn post from URL
• generate_article - Simple article from URL

📊 Scan & Source Management Tools

Monitor intelligence sources:

trigger_scan

Manually trigger intelligence scan.

Returns:

• Success status
• Scan ID
• Message

Example:

Use trigger_scan to start a new scan

get_scan_progress

Check scan status.

Returns:

• Is scanning (boolean)
• Progress percentage
• Current source
• Articles found count

Example:

Use get_scan_progress to see if scan is running

add_source

Add new RSS/HTML source to monitor.

Parameters:

• name (required) - Source name
• url (required) - RSS feed or website URL
• type (required) - rss or html

Example:

Use add_source to add TechCrunch RSS feed

📁 Resources

Read-only data access:

| Resource | Description | |----------|-------------| | verto://agents | Agent registry with configurations | | verto://articles/latest | Most recent scan results | | verto://articles/history | Historical scans (last 7 days) | | verto://sources | List of monitored sources | | verto://status | System status and last scan time |

Example:

Read the verto://agents resource to see agent registry

Usage Examples

Research → LinkedIn Post Workflow

1. Use execute_web_scout to research "AI pricing strategies for SaaS 2024"
2. Use execute_linkedin_writer with the research results
3. Use execute_voice_critic to validate the post

Full Multi-Agent Content Package

Use create_content_multi_agent with:
- inputType: "topic"
- sourceValue: "Modern data stack for marketing teams"
- outputType: "both"

Returns: LinkedIn post + article + scoring + voice check

Customize Agent Behavior

1. Use get_agent_prompt for agentName "linkedin_writer" to see current prompt
2. Use test_agent_prompt to try a more aggressive version
3. If satisfied, use update_agent_prompt to activate it

Environment Variables

| Variable | Required | Description | |----------|----------|-------------| | VERTO_API_URL | Yes | URL of your Verto Intel Bot instance | | VERTO_ADMIN_PASSWORD | Yes | Admin password for API authentication |

Response Format

All agent tools return formatted markdown for readability:

# Agent Execution: web_scout

✅ **Success**

## Output

{
  "sources": [...],
  "key_facts": [...],
  "content_angles": [...]
}

## Metadata

- **Execution Time:** 77000ms
- **Tokens Used:** 11934
- **Prompt Version:** 1

Architecture

Claude Desktop
    ↓
MCP Server (stdio)
    ↓
HTTP Client (VertoAPIClient)
    ↓
Main Bot API (/api/agents/*)
    ↓
Agent Registry → BaseAgent → LLM Providers

Key Components

• src/index.ts - MCP server main entry point (580 lines)
• src/types.ts - TypeScript interfaces for all tools (20+ interfaces)
• src/utils/api-client.ts - HTTP client for main bot communication
• src/tools/agent-tools.ts - Tool schema definitions (11 agent tools)
• src/tools/agent-handlers.ts - Tool handlers with markdown formatting

Development

# Clone the repo
git clone https://github.com/alexTNI/verto-intel-bot.git
cd verto-intel-bot/mcp-server

# Install dependencies
npm install

# Build
npm run build

# Watch mode
npm run dev

# Type checking
npm run type-check

Troubleshooting

"Connection refused"

• Ensure main bot is running on VERTO_API_URL
• Check firewall settings

"Authentication required"

• Set VERTO_ADMIN_PASSWORD environment variable
• Verify password matches main bot's ADMIN_PASSWORD

"Agent not found"

• Use list_agents to see available agents
• Verify main bot has initialized agents

Tools not appearing in Claude Desktop

• Rebuild MCP server: npm run build
• Check config path in claude_desktop_config.json
• Restart Claude Desktop
• Check logs: ~/Library/Logs/Claude/mcp*.log

License

MIT