Control every action, automate the routine, deliver what matters. A local gateway that governs what your agents can do — and a pipeline engine that collapses dozens of tool calls into one clean, validated answer.
Everything runs through MCPToolGate. The gate governs what agents can touch; the pipeline engine consumes that surface to collect and compress data. Run the gate alone as a security layer — the engine only ever works on top of it.
All requests pass through the same rules. What happens to a call depends only on the policy you set — the agent just sees the outcome.
Declarative schema, not code — an agent can author it from a skill and register it with one tool.
unmapped_policy: report.The engine runs the whole tool chain — fetch, merge, transform — without a single model token. The agent only pays for the final answer and for the sources that genuinely need reasoning. Measured on a live assistant across 18 sources and ~130 chat channels, same workload before and after.
Secure execution. Automated workflows. Smarter agents. Whether the agent calls a tool directly or runs a whole pipeline, it goes through the same governance layer.
One local binary. Pure Go, no runtime, cross-platform. Point it at your MCP servers, choose what's exposed, write a schema, and hand your agent signal instead of sprawl. MCPToolGate is in active development — join the waitlist to get early access first.