Token Diet
Route agent traffic through an MCP gateway that semantically prunes context so long coding sessions stay cheaper and within limits.
Overview
Token Diet is an MCP server for the Build phase that optimizes LLM tokens through a semantic pruning gateway.
What is this MCP server?
- LLM token optimization gateway exposed as remote SSE MCP (token-diet-mcp.vercel.app)
- Semantic pruning to trim low-value context instead of blunt character cuts
- Designed as an MCP layer agents can call or sit behind for leaner prompts
- Premium settlement via optional EIP-3009 payment-signature header
- Version 1.0.0 entry in the public MCP server schema catalog
- Server version 1.0.0
- One remote SSE URL: token-diet-mcp.vercel.app/api/mcp
- Implements semantic pruning as the stated optimization method
What problem does it solve?
Agents that resend bloated context on every turn silently inflate API bills and hit context ceilings on solo-builder budgets.
Who is it for?
Builders who run MCP-heavy agent workflows on large codebases and want a hosted optimization layer without building their own pruner.
Skip if: Users who need fully local, auditable token pipelines with no remote gateway or optional on-chain payment headers.
What do I get? / Deliverables
Registering Token Diet gives your stack an MCP-accessible pruning gateway so requests can carry leaner, semantically filtered context.
- MCP-connected token optimization gateway in your agent setup
- Semantically pruned context paths for supported agent queries
- Documented remote endpoint for team MCP configs
Recommended MCP Servers
Journey fit
Token dietting matters while you wire up agents and integrations during Build, before you scale usage in Grow and Operate. Semantic pruning gateways sit alongside other agent-tooling MCP servers that shape how models see your repo and chat history.
How it compares
Hosted LLM token gateway MCP, not a repository skill that teaches prompting tricks only.
Common Questions / FAQ
Who is Token Diet for?
Solo builders and small teams using MCP agents who want semantic context pruning to control LLM token spend.
When should I use Token Diet?
Use it while setting up agent-tooling in Build—or when Operate costs spike—from sessions that repeatedly send oversized context.
How do I add Token Diet to my agent?
Configure the remote MCP SSE URL https://token-diet-mcp.vercel.app/api/mcp in your client and supply payment-signature if premium settlement is required.