Irigate

Local MCP infrastructure

Share local MCP servers across your AI coding agents.

Irigate is a local MCP broker that lets AI coding agents connect to loopback Streamable HTTP while it supervises selected stdio MCP servers.

Measured at 20 identical-context Context7 clients: 95% fewer instances and 94.7% lower resident memory.

Get startedView source
Current local run
uv tool install --force --from . irigate
irigate --config profiles/mvp.yaml --check
irigate --config profiles/mvp.yaml --require-qualified-sharing

Before and after

One local broker, explicit upstream modes.

Before and after local MCP process routing

Before

Each AI coding agent launches its own stdio MCP server process.

Agentstdio serverAgentstdio serverAgentstdio server

After

Agents connect to one loopback Streamable HTTP broker; upstreams stay qualified-shared or session-isolated.

AgentsIrigate brokerqualified sharedsession isolated

Measured evidence

Process savings for identical-context Context7.

The benchmark in Benchmarks measured Context7 with identical credential-free, workspace-free contexts. It does not support a latency improvement claim because the calls were upstream-throttled.

5 clients80% fewer

Context7 instances and 79.0% lower resident memory in the measured identical-context run.

20 clients95% fewer

Context7 instances and 94.7% lower resident memory in the measured identical-context run.

LatencyNo claim

Throttled calls do not establish a latency improvement.

Differentiators

Built for local agent work, not a cloud control plane.

Workstation-local consolidation

Irigate runs beside local coding agents and targets duplicate stdio subprocesses on one workstation.

Fail-closed sharing

Upstreams are isolated by default. Shared mode requires explicit opt-in and a reviewed qualifier.

Metadata-only visibility

Reports track upstreams, tool names, durations, outcomes, and process counts without arguments or results.

How it works

Configure, start, connect.

  1. Configure a YAML profile with loopback listener settings and stdio upstreams.
  2. Start Irigate with --config; upstreams remain stopped until selected.
  3. Connect agents to a loopback URL and narrow access with tools= or upstreams=.

Quick start

A minimal local profile and selected client URL.

Install, validate, and start
uv tool install --force --from . irigate
irigate --config profiles/mvp.yaml --check
irigate --config profiles/mvp.yaml --require-qualified-sharing
Minimal profile shape
name: local-development
upstreams:
  context7:
    command: npx
    args: ["-y", "@upstash/context7-mcp"]
    idle_timeout_seconds: 300

Connect a client with the exact tool selector http://127.0.0.1:8765/mcp?tools=context7__resolve-library-id,context7__query-docs. See Getting started for the full path.

Compatibility and boundaries

Validated narrowly, isolated by default.

Questions

Check the visible FAQ before sharing a server.

Start with the FAQ for the sharing boundary, cloud-service answer, gateway comparison, and latency limitation.

Read the docs