Gateway overview
The VectorStep Gateway is a lightweight Python/FastAPI WebSocket gateway that runs AI agents with MCP tool access. It acts as an executor backend for VectorStep pipelines, providing an alternative to OpenClaw with support for multiple LLM providers and configurable MCP tool servers.
The gateway sits between VectorStep and your LLM providers. VectorStep sends an agent request over WebSocket; the gateway runs the full agentic loop (LLM calls, MCP tool execution, multi-turn conversation) and returns the final result. VectorStep never sees intermediate tool calls or thinking content — it gets one clean response.
This section is the Gateway’s own reference: configuration, agent authoring, the WebSocket and REST protocols, and day-two operations. If you just want the fastest path to a first pipeline running, see the site’s quick start instead — this page is the Gateway’s standalone introduction.
Quick Start
Section titled “Quick Start”export ANTHROPIC_API_KEY=sk-ant-...curl -sSL https://raw.githubusercontent.com/bantex01/VectorStep-Dist/main/install.sh | bashThe installer brings the Gateway up alongside VectorStep, seeds sample agents
into ~/.vectorstep/agents/, and handles the invoke token for you — the
Gateway mints it (plus a separate admin token, printed once for you to save)
on first boot, and the installer reads the invoke token out and writes it
into ~/.vectorstep/.env, so VectorStep is authenticated from the start with
nothing to copy and paste. See Gateway
authentication for what each token grants.
Then edit ~/.vectorstep/config/gateway.yaml with your LLM provider keys and
any MCP servers, add your own agent under ~/.vectorstep/agents/ (see
Creating agents), and apply the change:
cd ~/.vectorstepdocker compose up -ddocker compose logs -f gateway
# The invoke token, if you ever need it by handgrep VECTORSTEP_GATEWAY_TOKEN .envDirectory Structure
Section titled “Directory Structure”~/.vectorstep/├── docker-compose.yaml # managed by the installer, refreshed on re-run├── .env # keys, ports, image tag, gateway token — never overwritten├── config/│ ├── vectorstep.yaml # service config│ └── gateway.yaml # gateway config├── pipelines/ # host dirs, bind-mounted into the containers├── steps/└── agents/ # your agent definitionspipelines/, steps/, and agents/ are ordinary host directories — edit
them with any editor, the container sees the change immediately. The
database, artifacts, and logs live in a named Docker volume, not on the
host; reach them with docker compose exec, not a host path.
Where next
Section titled “Where next”- Configuration — every
config.yamlfield. - Providers — model routing and Azure OpenAI specifics.
- Creating agents —
agent.yaml,soul.md, hot reload. - WebSocket protocol — the
agentrequest/response contract. - REST API — health, agent management, MCP introspection.
- Operations — metrics, environment variables, performance notes.
- VectorStep integration — wiring the gateway into a pipeline.