I Built An Agentic Chatbot with Claude Code
“Hold up. You gave it tools and memory… and let it decide what to do?”
I did. And now it’s solving problems I didn’t even ask it to.
↳ GitHub 🔗 https://github.com/scarnyc/agentic-workflow
Last weekend, I went down the rabbit hole with Claude Code—and liked it so much that I ended up creating an agentic chatbot with a ton of features, including:
Web Search powered by Tavily Search
Code Execution in a secure Python env
Wikipedia Search
Vision Capabilities with PDFs
Short + Long-term Memory
Smart Writer with GPT-4o
Built-in reasoning + MCP registry in multi-turn conversations.
So I wired it up, gave it some tools, a personality, and a task:
“Help users solve complex problems with clarity—and autonomy.”
I used Claude Code to build it: Anthropic’s natural language code agent for developers who prefer to describe rather than debug—and it works fast!
It writes, edits, and explains Python, CSS, JS, and HTML as your one and only pair programmer right in the terminal.
It’s like having a full-stack ENG who literally develops everything to spec.
You can see me using it in the terminal window to convert LangGraph's tool implementations to MCP.
The result?
A chatbot that doesn’t just answer.
It thinks, reflects, and plans before using a tool or providing a response.
The Graph That Binds It
The chat agent is powered by LangGraph + Claude 4 Sonnet.
Here’s what stood out:
It reroutes when tools fail (self-healing workflows).
It queries memory before answering (contextual precision).
It thinks again before replying (meta-awareness).
So you’re collaborating with an agent that’s able to:
Break down complex problems into steps and workflows.
Decide which tools to use (and when).
Persist memory across sessions.
👀 What’s Next?
I’m thinking…
Let it schedule my meetings.
Or plug it into real workflows—support, strategy, research.
Agentic workflows are still early. But they feel like the missing link between LLMs and real utility.
This chatbot isn’t finished. It’s evolving…
Pre-MVP. Security features and UX enhancements are needed before I deploy it with Replit.
Check out the roadmap on GitHub ✌️
