Token-sparing retrieval
search_tools returns the top-k relevant tool cards per query (~200–800 tokens regardless of registry size). LLM context usage stays flat whether you manage 20 or 2,000 tools.
One endpoint for all your MCP servers. Clients retrieve only the LLM tools they actually need — optionally reranked by a decision model.

OptiGate is a self-hosted MCP (Model Context Protocol) gateway and server registry. Without a gateway, every AI agent needs every MCP server registered individually — and every tool schema lands in the LLM context, exploding token costs. OptiGate flips this around:
MCP Client ──▶ POST /mcp ──▶ OptiGate ──▶ managed MCP servers (HTTP/SSE/stdio)
search_tools + execute_tool onlysearch_tools(query, k) and gets the k most relevant tool cards.execute_tool(server_id, tool_name, args) — scope-checked, args-validated, audited.Head to installation to run it in five minutes, or read how decision-model reranking closes the paraphrase gap of keyword search.