Scrum Agent

Shikenso Logo

Introduction
Features
Architecture
Installation and Setup
Running Locally
Environment Variables
MCP Server Configuration
Wiki-based Skills
Usage Examples
Testing
Tracing with LangSmith
Roadmap
Contact

1. Introduction

Scrum Agent is an open-source, AI-powered Scrum Master that integrates Discord with Taiga project management. It uses a single ReAct agent powered by LangGraph and connects to external tools via the Model Context Protocol (MCP).

By mapping Discord channels to Taiga projects, Scrum Agent facilitates user story management through threaded discussions, automated stand-ups, and seamless synchronization between both platforms.

Join our Discord Server to test the Scrum Agent.

Project Goals & Audience

Scrum Agent is designed for small dev teams without a dedicated Scrum Master, but can also be used by larger teams to keep Discord conversations and Taiga boards in sync. The project is under active development — contributions and feedback are welcome.

2. Features

Discord Integration

Automated Discord thread creation for each Taiga user story
Semantic search over Discord message history (ChromaDB)
Daily stand-up posts at 08:00 (Berlin time)
Hourly Taiga-Discord synchronization
Smart message splitting for Discord's 2000-char limit

Taiga Project Management

Retrieve, create, and update user stories, tasks, and issues
Wiki page management (list, read, create, update with optimistic locking)
Sprint tracking and status updates
Watchers and assignee management

Wiki-based Skills

Dynamic system prompt extensions loaded from Taiga wiki pages
No code changes needed — create a wiki page with the skills- prefix and it is automatically discovered
TTL-based auto-refresh (default: 5 minutes) — update skills without restarting the bot
Graceful degradation when Taiga is unreachable

MCP Tool Servers

Taiga MCP — Full Taiga API access via langchain-taiga
GitHub MCP — Repository browsing, commits, branches (Docker)
Discord Chroma MCP — Semantic search over Discord messages
Discord API MCP — Direct Discord channel/message operations

Web Research

DuckDuckGo search, ArXiv papers, YouTube, Wikipedia

Multi-LLM Support

OpenAI (default: gpt-4o)
Anthropic Claude (claude-sonnet-4-5-20250929, etc.)
Ollama local models (ollama/llama3, etc.)

Stand-up Scheduling via Tags

| Tag on Taiga user story | Bot behaviour | |-------------------------|---------------| | daily stand-up | Stand-up every day | | weekly stand-up | Stand-up Mondays only | | no stand-up | No stand-up | | no tag | Mondays only (default) |

3. Architecture

Scrum Agent v2.0 uses a single ReAct agent (LangGraph create_react_agent) that connects to external tools via MCP (Model Context Protocol). Each MCP server runs as a subprocess and communicates via stdio.

Discord Bot (main_discord_bot.py)
  └── ScrumAgent (agent.py)
        ├── Skills Loader (Taiga wiki → system prompt, TTL-cached)
        ├── LangGraph ReAct Agent
        │     ├── LLM (OpenAI / Anthropic / Ollama)
        │     ├── System Prompt (base + loaded skills)
        │     └── Tools
        │           ├── MCP: Taiga (langchain-taiga)
        │           ├── MCP: GitHub (Docker container)
        │           ├── MCP: Discord Chroma (custom, ChromaDB)
        │           ├── MCP: Discord API (mcp-discord npm)
        │           ├── DuckDuckGo Search
        │           ├── ArXiv
        │           ├── YouTube Search
        │           └── Wikipedia
        └── Checkpointer (MongoDB or MemorySaver)

Key Files

| File | Description | |------|-------------| | scrumagent/agent.py | Core agent with MCP client and LLM factory | | scrumagent/main_discord_bot.py | Discord bot entry point | | mcp_servers/discord_chroma_server.py | Custom MCP server for semantic Discord search | | config/mcp_config.yaml | MCP server configuration | | config/taiga_discord_maps.yaml | Discord-to-Taiga project mappings | | pyproject.toml | Dependencies and project config |

4. Installation and Setup

Prerequisites

Python 3.10–3.12
uv (Python package manager)
Docker (for GitHub MCP server)
Node.js / npm (for Discord API MCP server)

Install Dependencies

git clone https://github.com/Shikenso-Analytics/ScrumAgent.git
cd ScrumAgent
git checkout claude/scrumagent-langgraph-mcp-bz738

# Install all dependencies including dev tools
uv sync --all-extras

Configure Environment

cp .env.example .env
# Edit .env with your credentials (see Section 6)

cp config/taiga_discord_maps.yaml.example config/taiga_discord_maps.yaml
# Edit the mapping file to match your Discord channels and Taiga projects

Discord Bot Setup

Go to the Discord Developer Portal
Create a new Application and go to the Bot tab
Enable Server Members Intent and Message Content Intent
In OAuth2, enable the bot scope with these permissions:
- View Channels, Send Messages, Read Message History
- Create Public/Private Threads, Manage Threads, Send Messages in Threads
- Manage Messages, Add Reactions
Use the generated OAuth2 URL to invite the bot to your server
Copy the bot token and add it to your .env file as DISCORD_TOKEN

5. Running Locally

Start the Discord Bot (production mode)

uv run python scrumagent/main_discord_bot.py

This starts the Discord bot, which will:

Connect to Discord and establish MCP server connections
Create/update threads for each active Taiga user story
Begin the daily stand-up and hourly sync tasks
Listen for messages in configured channels

Debug a Single Agent Invocation

To test the agent without the Discord bot, use a Python script:

uv run python -c "
import asyncio
from scrumagent.agent import ScrumAgent
from langchain_core.messages import HumanMessage

async def main():
    agent = ScrumAgent()
    await agent.start()
    print(f'Tools loaded: {len(agent.graph.get_graph().nodes)}')

    result = await agent.ainvoke(
        [HumanMessage(content='List all active user stories in project my-project-slug')],
        {'configurable': {'user_id': 'debug', 'thread_id': 'debug-session'}}
    )
    print(result['messages'][-1].content)

asyncio.run(main())
"

Test Individual MCP Servers

# Taiga MCP — should print FastMCP banner and start listening
uv run python -m langchain_taiga.mcp_server

# Discord Chroma MCP — starts and waits on stdio
uv run python mcp_servers/discord_chroma_server.py

# GitHub MCP (requires Docker + GITHUB_PERSONAL_ACCESS_TOKEN)
docker run -i --rm -e GITHUB_PERSONAL_ACCESS_TOKEN ghcr.io/github/github-mcp-server

# Discord API MCP (requires npm + DISCORD_TOKEN)
DISCORD_TOKEN=your_token npx -y mcp-discord

Test MCP Tool Loading (without credentials)

# Verify all 4 web tools load
uv run python -c "
from scrumagent.agent import _build_web_tools
tools = _build_web_tools()
print(f'{len(tools)} tools: {[type(t).__name__ for t in tools]}')
"

# Verify MCP config loads and env vars resolve
uv run python -c "
from scrumagent.agent import _load_mcp_config
config = _load_mcp_config()
for name, cfg in config.items():
    env_keys = list(cfg.get('env', {}).keys()) if 'env' in cfg else 'none'
    print(f'{name}: env={env_keys}')
"

# Verify LLM factory
uv run python -c "
from scrumagent.agent import _build_llm
llm = _build_llm()
print(f'LLM: {type(llm).__name__}')
"

6. Environment Variables

Copy .env.example to .env and fill in the values:

| Variable | Required | Description | |----------|----------|-------------| | SCRUM_AGENT_MODEL | No | LLM model name. Default: gpt-4o. Options: claude-sonnet-4-5-20250929, ollama/llama3 | | SCRUM_AGENT_TEMPERATURE | No | LLM temperature. Default: 0 | | DISCORD_TOKEN | Yes | Discord bot token | | DISCORD_GUILD_ID | Yes | Discord server (guild) ID | | DISCORD_THREAD_TYPE | No | public_thread or private_thread. Default: public_thread | | OPENAI_API_KEY | Yes | OpenAI API key (for embeddings and default LLM) | | GITHUB_PERSONAL_ACCESS_TOKEN | No | GitHub token for GitHub MCP server | | CHROMA_DB_PATH | No | ChromaDB storage path. Default: resources/chroma | | TAIGA_API_URL | Yes | Taiga API URL (e.g. https://api.taiga.io) | | TAIGA_URL | Yes | Taiga web URL (e.g. https://tree.taiga.io) | | TAIGA_TOKEN | Yes* | Taiga auth token (alternative to username/password) | | TAIGA_USERNAME | Yes* | Taiga username (alternative to

ScrumAgent

Install / Use

README