goclaw
Multi-agent AI gateway with teams, delegation & orchestration. Single Go binary, 11+ LLM providers, 5 channels.
claude mcp add --transport stdio nextlevelbuilder-goclaw docker run -i nextlevelbuilder/goclaw:latest \ --env PGHOST="HOST" \ --env PGUSER="USER" \ --env PGDATABASE="DBNAME" \ --env PGPASSWORD="PASSWORD" \ --env DATABASE_URL="postgres://USER:PASSWORD@HOST:5432/DBNAME" \ --env OTEL_EXPORTER_OTLP_ENDPOINT="http://localhost:4317"
How to use
GoClaw is a multi-tenant AI gateway written in Go that connects LLMs to tools, channels, and data. It orchestrates agent teams, inter-agent delegation, and quality-gated workflows across 13+ providers with strong security and observability. The MCP integration enables runtime-based tool access and streamable interactions via standard HTTP/WebSocket channels, making it possible to deploy a single binary that serves multiple tenants. Use the included HTTP API and WebSocket channels to manage agents, handoffs, and evaluator loops, while taking advantage of per-user workspaces stored in PostgreSQL and AES-256-GCM encrypted secrets.
How to install
Prerequisites:
- Docker installed on the host (recommended for quick setup)
- Optional: Go development environment if you prefer building from source
Option A: Run via Docker (recommended)
- Ensure Docker is running and you have access to a PostgreSQL instance (or use a managed DB).
- Pull and run the GoClaw image:
docker run -d --name goclaw
-e DATABASE_URL="postgres://USER:PASSWORD@HOST:5432/DBNAME"
-e PGHOST="HOST"
-e PGUSER="USER"
-e PGPASSWORD="PASSWORD"
-e PGDATABASE="DBNAME"
-e OTEL_EXPORTER_OTLP_ENDPOINT="http://localhost:4317"
nextlevelbuilder/goclaw:latest - Ensure the container can reach the PostgreSQL database and that firewall rules allow traffic.
Option B: Build from source (Go)
- Prerequisites: Go 1.25+ installed, PostgreSQL, and network access to the DB.
- Clone the repository and build: git clone https://github.com/nextlevelbuilder/goclaw.git cd goclaw go build -trimpath -o goclaw
- Run the binary with environment variables configured for your PostgreSQL instance, for example: DATABASE_URL=postgres://USER:PASSWORD@HOST:5432/DBNAME ./goclaw
- Verify the service starts within a few seconds and is accessible via its HTTP/WebSocket endpoints.
Additional notes
Environment and configuration tips:
- The GoClaw deployment relies on a multi-tenant PostgreSQL backend. Ensure each user has an isolated schema/workspace and that AES-256-GCM-based encryption keys are managed securely.
- If using Docker, expose the necessary ports for HTTP/WebSocket API and OTLP if you enable OpenTelemetry tracing.
- For observability, you can enable OTLP exporting by setting OTEL_EXPORTER_OTLP_ENDPOINT to your collector endpoint.
- When upgrading, prefer image_TAG updates for Docker deployments to minimize surface changes. Verify that the PostgreSQL schema migrations are applied during startup.
- Common issues: network access from the MCP gateway to PostgreSQL, invalid DATABASE_URL formats, and insufficient DB user permissions. Check logs for authentication failures or SSL configuration mismatches.
Related MCP Servers
semantic-router
System Level Intelligent Router for Mixture-of-Models at Cloud, Data Center and Edge
go-utcp
Official Go implementation of the UTCP
local-skills
Universal MCP server enabling any LLM or AI agent to utilize expert skills from your local filesystem. Reduces context consumption through lazy loading. Works with Claude, Cline, and any MCP-compatible client.
mcp-client-gen
Turn any MCP server into a type-safe TypeScript SDK in seconds - with OAuth 2.1 and multi-provider support
nucleus
The Sovereign Agent Control Plane — One brain for Cursor, Claude, and Windsurf.
agi -agent
A modular AGI agent framework based on MCP (Multi-Context Processing), inspired by Manus, with ChatGPT-style LLM integration and task control.