MCPsServers by JosueSay - MCP Server

Claude Console + MCP (FEL + Overwatch)

This project implements a console chatbot that connects to Anthropic Claude via API and integrates multiple Model Context Protocol (MCP) servers. The chatbot can maintain conversational context, automatically decide when to use tools, and log all interactions in structured JSONL files.

It includes a local FEL MCP that validates and renders Guatemalan electronic invoices (FEL) in PDF with watermark, and a remote Overwatch MCP (HTTP, cloud-deployed) that allows querying player statistics.

Monorepo notice: This repository consolidates two implemented codebases: the chatbot (CLI/UI) and the local FEL MCP. Additionally, it includes an Overwatch MCP (HTTP, Cloud Run) and links a reference repository for connection patterns.

🔗 Related Repositories

Chatbot (CLI / UI) — Integrated in this monorepo.
FEL MCP (Local) — Integrated in this monorepo.
Reference: OpenAI Chat API Example — Reference only (connection patterns).

✨ Features

Connects to Anthropic Claude API (LLM).
Maintains conversational context throughout the session.
Supports manual and automatic tool usage.
Integrated tools:
- FEL MCP (local):
  - fel_validate: validates XML totals (subtotal, 12% VAT, total).
  - fel_render: generates PDF invoices with watermark from XML.
  - fel_batch: processes multiple XML invoices → PDFs + manifest.json.
- Overwatch MCP (HTTP, cloud):
  - ow_get_player_summary: summary of player stats (winrate, KDA, damage, healing).
  - ow_get_player_stats: full stats with tags (by role, hero, mode).
Session logs saved in structured JSONL.
Sandbox paths (ALLOWED_ROOTS) to prevent unsafe file access.

⚙️ Requirements

Python 3.12
Ubuntu 22.04 (WSL) or Linux environment
Virtual environment (recommended)

🔧 Installation

When starting the chatbot, you can now run multiple MCP servers in parallel. This allows combining local tools (e.g., FEL) with external integrations such as Overwatch (HTTP cloud), Filesystem, and GitHub.

Environment Variables

Edit your .env with the following variables:

# ---- Keys ----
ANTHROPIC_API_KEY=your_key
ANTHROPIC_MODEL=claude-sonnet-4-20250514 # optional, other models: https://docs.anthropic.com/en/api/models-list
GITHUB_PERSONAL_ACCESS_TOKEN=your_token # Personal Access Token (fine-grained or classic)

# ---- MCPS ----
MCP_FEL_CMD="/ABSOLUTE/PATH/venv/bin/python /ABSOLUTE/PATH/servers/fel_mcp_server/server_stdio.py"
MCP_URL=https://ow-mcp-server-XXXXXX.run.app/mcp # Your Overwatch MCP cloud URL

# Multiple MCPs (comma-separated)
# Recommended order: FEL (local), Filesystem (Docker), GitHub (Docker)
MCP_CMDS="/ABSOLUTE/PATH/venv/bin/python /ABSOLUTE/PATH/servers/fel_mcp_server/server_stdio.py,docker run --rm -i -v /ABSOLUTE/PATH/data/testing:/data node:22 npx -y @modelcontextprotocol/server-filesystem /data,docker run --rm -i -e GITHUB_PERSONAL_ACCESS_TOKEN node:22 npx -y @modelcontextprotocol/server-github"

# ---- Chat Config ----
LOG_SESSION=/ABSOLUTE/PATH/TO/REPO/data/logs/sessions
LOG_RPC=/ABSOLUTE/PATH/TO/REPO/data/logs/rpc
ROUTER_DEBUG=0 # 1: active | 0: inactive

GitHub MCP

To use the GitHub MCP, you need to:

Create a personal token (fine-grained or classic).

Clone the official MCP servers repository to an absolute path:

git clone https://github.com/modelcontextprotocol/servers.git

Replace /ABSOLUTE/PATH/servers in your .env with the actual cloned path.

This way, the chatbot can launch and orchestrate multiple MCPs in parallel.

🚀 Usage

Start the chatbot from the project root:

python apps/cli/chat.py

You will see something like:

───────────────────── Claude Console + MCP (FEL + Overwatch) ──────────────────────
commands: /exit | /clear | /tools | /paste | fel_validate <xml> | fel_render <xml>    | fel_batch <dir_xml>
You ›

FEL Example

You › /tools
MCP Tools
name         | description                                  | required args
-------------|----------------------------------------------|---------------
fel_validate | Validate FEL XML totals and required fields  | xml_path
fel_render   | Generate PDF with watermark from FEL XML     | xml_path
fel_batch    | Process FEL XML directory → manifest         | dir_xml

You › puedes validar precios de data/xml/factura.xml?
La validación de la factura fue exitosa. Los totales calculados son:

 • Subtotal: Q8,010.59  
 • IVA (12%): Q961.27  
 • Total: Q8,971.86  

No se encontraron errores en los precios ni en los campos requeridos.

Overwatch Example (HTTP)

You › resume Overwatch stats for player YSE#11202 using ow_get_player_summary
Overall Performance:

 • 4,263 games played with a 52.8% winrate (2,251 wins, 2,012 losses)
 • 608 hours total playtime
 • 3.03 KDA ratio with 15.07 average eliminations per game

Role Distribution:

 • Tank: 1,699 games (51.85% winrate, 3.27 KDA)
 • Damage: 1,241 games (53.42% winrate, 3.08 KDA)
 • Support: 1,323 games (53.44% winrate, 2.64 KDA)

Top Heroes by Games Played:

 1 Doomfist - 523 games (47.8% winrate)
 2 Lucio - 332 games (58.73% winrate)
 3 Junkrat - 289 games (56.06% winrate)
 4 Reinhardt - 229 games (60.26% winrate)
 5 Cassidy - 203 games (55.17% winrate)

The player shows solid performance across all roles with slightly better winrates on damage and support heroes compared to tank. Lucio and Reinhardt appear to be among their strongest heroes with high winrates.

With absolute paths configured for official servers, you can now access more tools: FEL + Overwatch (cloud) + Filesystem + GitHub.

MCP Tools in action

🗂 Project Structure

.
├── apps/
│   ├── cli/
│   │   └── chat.py             # Chatbot CLI frontend
│   └── ui/                     # Pending: possible UI frontend
├── core/
│   ├── engine.py               # Orchestration (Claude + MCP)
│   ├── mcp_stdio.py            # JSON-RPC client (MCP via stdio/local)
│   ├── mcp_http.py             # JSON-RPC client (MCP via HTTP/remote)
│   └── settings.py             # Configuration and environment variables
├── servers/
│   ├── fel_mcp_server/         # FEL MCP (validate, render, batch)
│   └── ow_mcp_server/          # Overwatch MCP (HTTP + stdio)
│       ├── config.py           # Constants (platforms, modes, roles)
│       ├── ow_api.py           # Overfast API calls (stats and summary)
│       ├── server_http.py      # MCP server via HTTP (cloud deployment)
│       └── server_stdio.py     # MCP server via stdio (for local use)
├── data/
│   ├── logos/                  # Company logos
│   ├── xml/                    # Example FEL invoice XMLs
│   ├── out/                    # Output PDFs
│   └── testing/                # Test files (filesystem MCP)
├── logs/sessions/              # Session logs (.jsonl)
├── assets/
│   ├── fonts/                  # Custom fonts (Montserrat, Roboto Mono)
│   └── images/                 # Icons (phone, email, web)
└── docs/                       # Project and MCP documentation

Workflow

flowchart LR
  subgraph User["CLI / UI"]
    U["Terminal<br/>apps/cli/chat.py"]
  end

  subgraph Core["Core (chatbot engine)"]
    E["engine.py<br/>(Claude + MCP router)"]
    S1["mcp_stdio.py"]
    S2["mcp_http.py"]
    CFG["settings.py<br/>(.env)"]
    LOG["logs/sessions/*.jsonl"]
  end

  subgraph LLM["Anthropic Claude API"]
    C["Claude<br/>(model)"]
  end

  subgraph MCPs_Local["Local MCPs (STDIO)"]
    FEL["fel_mcp_server/server_stdio.py<br/>FEL: validate/render/batch"]
    FS["docker server-filesystem<br/>(@mcp/server-filesystem)"]
    GH["docker server-github<br/>(@mcp/server-github)"]
  end

  subgraph MCPs_Remote["Remote MCPs (HTTP)"]
    OW["ow_mcp_server/server_http<br/>Cloud Run URL /mcp<br/>ow_get_player_*"]
  end

  subgraph RepoData["Local resources"]
    ASSETS["assets/<br/>fonts, images"]
    DATA["data/<br/>xml, out, testing"]
  end

  U --> E
  E --> C

  E -. "tools:list/call" .-> S1
  E -. "tools:list/call" .-> S2

  S1 --> FEL
  S1 --> FS
  S1 --> GH

  S2 --> OW

  E <--> LOG
  FEL -. "uses" .-> ASSETS
  FEL <--> DATA
  FS -. "mounts" .-> DATA

📝 Logs

Every session is automatically logged in logs/sessions/ as JSONL.

Example log entries:

{
  "type": "mcp",
  "op": "tools/list",
  "result": { ... }
}
{
  "type": "llm_auto",
  "model": "claude-sonnet-4-20250514",
  "input": "puedes validar precios de data/xml/factura.xml?",
  "output": "La validación de la factura fue exitosa..."
}

These logs allow you to trace tool usage, LLM decisions, and outputs.

🔒 Security Notes

ALLOWED_ROOTS defines directories where MCP tools can access files (data/xml, data/out, data/logos by default).
Any path outside of these roots will be blocked for safety.
Logs may contain sensitive invoice data -> review before sharing.

🖥️ Using with Claude Desktop + MCP

You can also run the FEL server directly inside Claude Desktop via the Model Context Protocol (MCP).

1. Install Claude Desktop

Download and install Claude from here: 👉 https://claude.ai/download

⚠️ Claude Desktop is available for Windows/macOS.

Our FEL server was designed to run inside WSL (Ubuntu).

2. Configure MCP

Open Claude Desktop and go to: File -> Settings -> Developer -> Edit Config

This will open the configuration folder. Edit the file claude_desktop_config.json and add:

{
  "mcpServers": {
    "FEL": {
      "command": "wsl.exe",
      "args": [
        "-e",
        "<absolute_path>/venv/bin/python",
        "<absolute_path>/servers/fel_mcp_server/server_stdio.py"
      ]
    }
  }
}

🔑 Note: Replace <absolute_path> with the full absolute path inside WSL, e.g.: /mnt/d/repositorios/UVG/2025/MCPsServers

3. Restart Claude Desktop

After saving the config file, restart Claude Desktop from PowerShell:

Stop-Process -Name "Claude" -Force; Start-Process "<absolute_path>\Claude.exe"

Here <absolute_path>\Claude.exe should be replaced with the full path to your Claude installation, for example: C:\Users\<username>\AppData\Local\AnthropicClaude\Claude.exe

📝 Prompts for Testing

These are the prompts used for testing the chatbot. They differ in language to evaluate how well the chatbot adapts to Spanish and English inputs.

Normal

¿Quién es Tom Cruise?

He's dead?

FEL

Can you generate a PDF of data/xml/factura.xml?

¿Puedes verificar los precios de data/xml/factura.xml?

¿Cuánto sería el total si lo multiplicamos por 2?

FileSystem

Lista todos los archivos que hay en el directorio /data

Read the complete contents of the file /data/file1.txt

GitHub

Busca información del repositorio JosueSay/SchedulerSim y muéstrame su README.

Get the list of commits from the main branch of the repository JosueSay/SchedulerSim.

MCP Remote OW

Dame un resumen de las estadísticas del jugador YSE#11202 en Overwatch (plataforma PC), incluyendo winrate, KDA, daño y sanación usando ow_get_player_summary.

Show me the detailed labeled career stats for the player YSE#11202 on PC in quickplay mode, using ow_get_player_stats.

JosueSay/MCPsServers