Ayush Shivani

Arka by KenisLabs - Enterprise MCP gateway

by
MCP breaks when you try to use it at scale. Context gets too big with many tools. Setup is messy with many servers. Security is missing. Arka fixes this with one gateway. One setup. One token with SSO. Tool filtering keeps context small. User rules and logs included. Open source and cloud options. Try it here: https://www.kenislabs.com

Add a comment

Replies

Best
Ayush Shivani
Maker
📌
We are excited to share Arka. We built it because MCP works well in demos but fails in real teams. Context explodes. Setup gets messy. Security is missing. Arka solves these issues with one gateway. All MCP servers connect once. Users get one token with SSO. Uses dynamic tool calling hence only sees tools it needs, so context stays small. Teams get logs and clear user rules. We have an open source version if you want to self host. Repo: https://github.com/KenisLabs/ark... We also have a cloud version. https://www.kenislabs.com If your team wants to adopt MCP, feel free to reach out. I am happy to chat and share what we learned. Calendly: https://calendly.com/ayushshivan... Thank you for checking it out.
Masum Parvej

@ayush_shivani Cool that you offer both open source and cloud. Do you expect most teams to start self‑hosted first

Ayush Shivani

@masump Great question.
I would expect most teams start with the cloud version because it removes setup work.
Self-hosting is the way to go once they want full control or need to plug into internal systems.

sarvesh mehta

We are in process of adding more MCP servers to extend functionality, and also bring unified secured authentication for AI agents.

Chilarai M

Congrats on the launch!

Ayush Shivani

@chilarai Thank you!

Harkirat Singh

Nice approach @ayush_shivani @sarvesh_mehta . I’m curious how well Arka keeps context lean in long-running workflows - does the dynamic tool calling prevent the usual context bloat teams run into?

Ayush Shivani

@harkirat_singh3777 Thanks!

Our approach does prevent the context bloat that happens when all tools from all MCP servers are kept in memory. We only provide the tools needed for the workflow at that moment through dynamic selection.
Also we do code execution, and the LLM parses the output to remove unnecessary details, which helps reduce context even further for long-running workflows.