Skip to content

mcp_server

Directory actions

More options

Directory actions

More options

Latest commit

 

History

History

mcp_server

README.md

Company Intelligence MCP Server

An MCP server that generates company intelligence reports. Accessible via REST API using FastMCP.

Quick Start

# 1. Navigate to directory
cd examples/mcp_server

# 2. Install dependencies
uv add fastmcp tavily-python

# 3. Set environment variables
cp .env.example .env
# Set TAVILY_API_KEY and OPENAI_API_KEY in .env

# 4. Start server
uv run python server.py

# 5. Run workflow only (without server)
uv run python workflow.py "Treasure Data"

# 6. Client example
uv run python client_example.py "Treasure Data"

Overview

This MCP server executes the following workflow using a company name as input to generate an intelligence report for business meeting preparation:

Search → Curate → Write → Critique → Output
       ↑                    ↓
       └──── Feedback Loop ─┘

Workflow Details

  1. Search: Collect information from multiple sources using Tavily API

    • Company news
    • Industry trends
    • Company profile
    • Competitor information

  2. Curate: Filter and organize search results with LLM

    • Relevance scoring
    • Deduplication
    • Category classification

  3. Write: Generate intelligence report with LLM

    • Executive summary
    • Detailed sections
    • Key talking points for meetings

  4. Critique: Evaluate the report with LLM

    • Accuracy, completeness, relevance, structure, source quality
    • Sends back to Write if criteria not met (max 3 iterations)

  5. Output: Export final report in Markdown format

Environment Setup

1. Create .env file

cd examples/mcp_server
cp .env.example .env
# Edit .env file to set your API keys

Required Environment Variables

TAVILY_API_KEY=your-tavily-api-key
OPENAI_API_KEY=your-openai-api-key

Optional (Langfuse Tracing)

LANGFUSE_PUBLIC_KEY=your-langfuse-public-key
LANGFUSE_SECRET_KEY=your-langfuse-secret-key
LANGFUSE_HOST=http://localhost:3000
ENABLE_TRACING=true

Other Settings

GRAFLOW_LLM_MODEL=gpt-5-mini      # Default model
WRITER_MODEL=gpt-5-mini           # Model for Writer
CRITIQUE_MODEL=gpt-5-mini         # Model for Critique
MAX_CRITIQUE_ITERATIONS=3          # Maximum revision count
MAX_SEARCH_RESULTS=10              # Maximum search results
MCP_SERVER_HOST=0.0.0.0            # Server host
MCP_SERVER_PORT=9100               # Server port

Starting the Server

Server Startup

# Navigate to directory
cd examples/mcp_server

# Start server
uv run python server.py

# Or start with uvicorn
uv run uvicorn server:app --host 0.0.0.0 --port 9100

Run Workflow Only (without server)

cd examples/mcp_server
uv run python workflow.py "Treasure Data"

API Endpoints

REST API

Endpoint Method Description
/health GET Health check
/config GET Get current configuration
/mcp POST MCP SSE endpoint

MCP Tools

generate_company_intelligence

Generate a comprehensive intelligence report for a company.

Input:

{
  "company_name": "Treasure Data",
  "enable_tracing": true
}

Output:

{
  "company_name": "Treasure Data",
  "report_markdown": "# Treasure Data Intelligence Report\n...",
  "executive_summary": "...",
  "key_takeaways": ["Point 1", "Point 2"],
  "sections": [...],
  "sources_count": 15,
  "iterations": 2,
  "generated_at": "2024-01-15T10:30:00",
  "critique_score": 0.85
}

search_company_news

Search for the latest news about a company (lightweight version).

Input:

{
  "company_name": "Salesforce",
  "max_results": 10
}

search_industry_trends

Search for industry trends.

Input:

{
  "company_name": "Sony",
  "industry": "Electronics",
  "max_results": 10
}

Client Usage

Python Client Example

cd examples/mcp_server

# After starting the server, in another terminal
uv run python client_example.py "Treasure Data"

# News search only
uv run python client_example.py "Salesforce" --action news

# Industry trends search
uv run python client_example.py "Sony" --action trends

Using curl

# Health check
curl http://localhost:9100/health

# Generate report (via SSE)
curl http://localhost:9100/mcp

Using httpx

import httpx

# Health check
response = httpx.get("http://localhost:9100/health")
print(response.json())

# Check configuration
response = httpx.get("http://localhost:9100/config")
print(response.json())

Claude Code MCP Setup

You can use this MCP server from Claude Code.

MCP Installation Scopes

MCP servers can be configured in three scopes:

Scope Storage Location Use Case
local ~/.claude.json (under project path) Personal use, current project only (default)
project .mcp.json at project root Team sharing (commit to version control)
user ~/.claude.json Available across all projects

Method 1: CLI Installation (Recommended)

First, start the server:

cd examples/mcp_server && uv run python server.py

In another terminal, register MCP:

# Local scope (default) - current project only
claude mcp add --transport http company-intel http://localhost:9100/mcp

# Project scope - shared with team (saved to .mcp.json)
claude mcp add --transport http --scope project company-intel http://localhost:9100/mcp

# User scope - available across all projects
claude mcp add --transport http --scope user company-intel http://localhost:9100/mcp

Method 2: Project Configuration File (Team Sharing)

Create .mcp.json in your project root (commit to version control):

{
  "mcpServers": {
    "company-intelligence": {
      "command": "uv",
      "args": ["run", "python", "server.py"],
      "cwd": "./examples/mcp_server",
      "env": {
        "TAVILY_API_KEY": "${TAVILY_API_KEY}",
        "OPENAI_API_KEY": "${OPENAI_API_KEY}"
      }
    }
  }
}

Environment variables can be referenced using ${VAR} format (actual values taken from user's environment).

Managing MCP Servers

# List registered servers
claude mcp list

# Get server details
claude mcp get company-intel

# Remove a server
claude mcp remove company-intel

Verification

Restart Claude Code and use the /mcp command to check connection status.

Langfuse Tracing

Workflow execution can be traced with Langfuse:

  1. Create an account at Langfuse Cloud
  2. Create a project and obtain API keys
  3. Set environment variables in the .env file
  4. After starting the server, view traces in the Langfuse dashboard

Traces include:

  • Total workflow execution time
  • Execution time for each task (search, curate, write, critique)
  • LLM call details (prompts, responses, token counts)
  • Errors and stack traces

Project Structure

examples/mcp_server/
├── __init__.py              # Module initialization
├── config.py                # Configuration management (dotenv support)
├── workflow.py              # Graflow workflow definition
├── server.py                # FastMCP server
├── client_example.py        # Client example
├── .env.example             # Environment variable template
├── claude_mcp_config.json   # Claude Code MCP config example
├── README.md                # This file
├── README_ja.md             # Japanese README
└── agents/
    ├── __init__.py
    ├── search.py        # Search agent (Tavily)
    ├── curator.py       # Curation agent (LLM)
    ├── writer.py        # Writer agent (LLM)
    └── critique.py      # Critique agent (LLM)

Dependencies

graflow
fastmcp
tavily-python
litellm
python-dotenv
httpx
uvicorn
pydantic

Installation:

uv add fastmcp tavily-python python-dotenv

License

Apache License 2.0