What is Backpressure?
A flow control mechanism where the proxy signals the agent to slow down when downstream MCP servers are overloaded. Backpressure prevents cascading failures by propagating load information upstream to the agent, allowing the system to degrade gracefully.
WHY IT MATTERS
MCP servers have finite capacity. A database server can handle only so many concurrent queries. A code execution server has limited CPU. When agents push more work than the server can handle, responses slow down, timeouts increase, and eventually the server fails — taking all connected agents down with it.
Backpressure is the engineering discipline of pushing load information upstream. Instead of accepting all requests and letting the server drown, the proxy monitors server health indicators — response latency, error rates, queue depth — and signals agents to slow down before the server reaches breaking point.
In distributed systems, backpressure prevents cascading failures. Without it, an overloaded server causes agents to timeout and retry, creating even more load. With backpressure, agents receive early warning and reduce their request rate, allowing the server to recover. This is especially critical in multi-agent deployments where dozens of agents may share a single MCP server.
HOW POLICYLAYER USES THIS
Intercept monitors upstream MCP server health and can apply backpressure by throttling or denying tool calls when the server shows signs of overload. This is configured in the YAML policy as health-based rules — when response latency exceeds a threshold or error rates spike, Intercept automatically reduces throughput to protect the downstream server.