{"id":46942466,"title":"Show HN: MCPlexor – MCP multiplexer that cuts agent context usage by 95%","points":1,"user":"arustagi","time":1770621435,"time_ago":"8 hours ago","type":"link","content":"

I built MCPlexor to solve a token waste problem I kept running into with MCP-based agents.

The Problem:\nMCP (Model Context Protocol) is great for giving LLMs access to external tools. But if you connect multiple servers (GitHub, Linear, Postgres, Slack), you end up with 40-50k tokens of tool definitions injected into every request – before the agent even does anything.

On a 200k context model, that's 25% gone. On smaller models, it's worse. And most runs only use 1-2 tools.

The Solution:\nMCPlexor sits between your agent and your MCP servers. Instead of loading all tool definitions upfront:

Agent asks for a capability ("create an issue")\nMCPlexor routes to the right server using semantic matching\nOnly relevant tools get exposed\nResult: ~500 tokens overhead instead of ~20k.

Technical Details:\n- Written in Go, single binary, no runtime deps\n- Supports both stdio and HTTP transports\n- Stores credentials in OS keychain (macOS Keychain, Windows Credential Manager, Linux keyring via zalando/go-keyring)\n- Works with Claude Desktop, Cursor, Augment Code, or any MCP-compatible client

The routing logic can run entirely locally using Ollama. No API calls, no cost, works offline.

Business Model (for transparency):\n- Local/Ollama users: completely free\n- Cloud tier (on waitlist): we run inference on efficient small models instead of Opus/Pro, pass on savings, take a margin\nThe bet is that routing is a narrow enough task that a fine-tuned 7B model does it as well as a frontier model. Early testing suggests this works.

CLI uploads: https://github.com/arustagi101/mcplexor\nInstall: `curl -fsSL https://mcplexor.com/install.sh | bash`\nHappy to answer questions about the architecture, the routing approach, or anything else.","url":"https://www.mcplexor.com","domain":"mcplexor.com","comments":[],"comments_count":0}