This is a submission for the Notion MCP Challenge
What I Built
Every developer knows the high cost of context switching. When you are in a deep state of flow, stepping away from your IDE to answer a "quick question" in Slack or to manually log a bug report into a sprint tracker can cost you 20 minutes of focus. I wanted to build a system that protects the developer's most valuable asset: their attention.
I built Synapse, an autonomous, AI-powered "Focus Mode" agent that acts as an intelligent shield between a developer's deep work and their team's Slack channel.
When a developer runs the custom CLI command (npm run focus), Synapse boots up and begins listening to incoming Slack messages via a secure webhook. Instead of just forwarding notifications, Synapse actively intercepts them and uses Groq's lightning-fast Llama-3.3-70b model to determine the user's intent. From there, it takes autonomous action based on three core workflows:
- Intelligent Issue Triage: If a teammate reports a bug or a system error in Slack, Synapse recognizes the intent. It autonomously structures the unstructured Slack message and uses the Notion API to write a fully populated bug ticket directly into the engineering team's Notion Sprint Tracker.
- Context-Aware Q&A (RAG): If a teammate asks a technical question or asks for API specifications, Synapse does not interrupt the developer. Instead, it queries the company's Notion API documentation, reads the database properties to find the correct context, synthesizes a human-friendly answer, and replies directly in the Slack thread.
- Silent Auditing: To ensure the developer never misses critical context while they are heads-down, Synapse logs every action it takes (both the trigger message and its own response) into a private "Synapse Logs" Notion database. This creates a complete, automated paper trail for the developer to review once they exit focus mode.
Synapse proves that AI does not just have to be a chat interface; with the right tools, it can be a fully autonomous teammate that handles the operational friction of software engineering.
I built Synapse, an autonomous, AI-powered "Focus Mode" agent. Synapse acts as a shield between a developer's deep work and their team's Slack channel.
When activated via a custom CLI command (npm run focus), Synapse listens to incoming Slack messages. It uses Groq's blazing-fast Llama-3.3-70b model to determine the user's intent. If a teammate reports a bug, Synapse autonomously structures the data and writes a ticket directly to our Notion Sprint Tracker. If they ask a technical question, it searches our Notion API documentation, synthesizes a friendly answer, and replies in Slack. Finally, it logs every action it takes into a background Notion database so I have a complete paper trail of what the AI handled while I was coding.
Video Demo
Show us the code
Synapse: The Autonomous Developer Focus Shield
Demo Video
Overview
Context switching is a developer's worst enemy. When an engineer is in deep focus, constantly switching to Slack to answer documentation questions or manually log bug reports into a sprint tracker shatters productivity.
Synapse is an autonomous, AI-powered agent designed to act as a shield between a developer's deep work and their team's Slack channel. Activated via a custom CLI, Synapse intercepts incoming Slack messages, determines the user's intent using Groq's Llama 3 model, and takes direct action in Notion using the Model Context Protocol (MCP).
Features
- Automated Bug Tracking: If a teammate reports a bug in Slack, Synapse autonomously structures the data and writes a new ticket directly into the engineering team's Notion Sprint Tracker.
- RAG Documentation Search: If a teammate asks a technical question, the AI queries the Notion API to search the company docs, synthesizes a human-friendly answer…
How I Used Notion MCP
The Model Context Protocol (MCP) was the absolute game-changer for this architecture. Instead of writing custom API wrappers and complex logic to map Groq's tool-calling outputs directly to Notion's API, I used the MCP Notion Server (@ramidecodes/mcp-server-notion).
By bridging Groq's LLM with MCP, Synapse dynamically selects and executes these tools:
-
create-page: Used to dynamically generate new database rows in my Sprint Tracker (for bugs) and my Synapse Logs (for background auditing). MCP handled the strict payload formatting Notion requires for Database parent IDs and Rich Text properties. -
search: Used to create a fast RAG (Retrieval-Augmented Generation) pipeline. When a user asks about API Rate Limits, Synapse calls thesearchtool, reads the properties from my Notion Documentation database, and feeds that context back to the LLM to summarize the answer.
MCP turned a complex, multi-step API integration into a clean, unified toolset that the LLM could execute flawlessly in real-time.
Top comments (19)
Nice submission overall. If I had to point out the top 5 high-level things to fix first, they would be:
Secure the Slack webhook
Right now the endpoint appears to trust incoming requests too much. Slack signature verification should be added so random external requests cannot trigger bot actions.
Handle duplicate Slack events
Slack can resend the same event, which could lead to duplicate replies, duplicate logs, or even duplicate sprint tickets. Add deduplication using the event ID.
Do not rely only on the model for routing
The prompt says the assistant “must” use certain tools, but with
tool_choice: "auto"that is still not guaranteed. Important actions like doc search vs bug ticket creation should be enforced in code, not only in prompting.Improve failure handling
If tool arguments fail to parse, MCP breaks, or Slack posting fails, the bot may silently do nothing. It needs clearer fallback behavior and better separation between critical actions and optional logging.
Tighten control over Notion writes
Creating tickets and logs directly from Slack messages is powerful, but also risky. I would add stronger validation, permission checks, and limits so users cannot easily spam or create noisy data in Notion.
Interesting prototype, but these changes would make it much safer and more production-ready.
Thank you for taking the time to review the architecture and for the detailed, actionable feedback. Upgrading the webhook security, handling event deduplication, and enforcing the tool routing in the code are excellent points for scaling this from a prototype into a production environment. I am adding these five items directly to the roadmap for V2. I really appreciate the insights
Really cool project, I loved the CLI interface, I think both of us found Synapse is a great name for a project. My project is also called Synapse, it is a chat with a graph memory, nothing related to your topic, but I found it funny that both chose the same name, and it works for both contexts.
Great project, congrats
Haha, great minds think alike! It really is the perfect name for these kinds of AI projects. Thank you for the kind words on the CLI - I spent a lot of time trying to get that developer experience right.
I am definitely going to check out your version of Synapse. Building a chat with graph memory sounds like a fascinating architecture, and as a student, I'm always eager to see how senior engineers structure things under the hood. Really appreciate the support!
good implementation i must say.was wondering what would have happened if a certain question or more is not in the api documentation
Great question,for preventing the AI from hallucinating, i made a strict system in which if its unable to find the query, it gracefully admits it searched the notion documentation but couldnt find details on that topic.For future V2,would love to add further functionality:)
kinda agreed,rather than providing a wrong answer its kinda good it says that for a starter program.regardless,great work,best wishes for future
thanks a bunch :)
Nice implementation of Notions MCP in everyday life,great work
thanks :)
VERY INFORMATIVE N PRACTICAL PROJECT,KEEP UP!
thanks a lot :)
Hey everyone! This is my submission for the Notion MCP challenge. What is your biggest distraction when you are trying to code?
AS A DEV, I PERSONALLY FEEL A NEED FOR SUCH A TOOL THAT CAN HANDLE WEIGHT OFF MY SHOULDERS
Someone handling your communication while you work does seem like lifting a weight off your shoulders,hope to improve it further,and thanks again :)
I was recently looking towards something that can do such tasks,maybe further add an ability that puts up a notification if there is an emergency
That functionality can be misused by people to convey any sort of message,even if there's no emergency,breaking the whole purpose of the program,hope that makes it clear :)
Good experience.
Thanks :)
Some comments may only be visible to logged-in visitors. Sign in to view all comments.