Skip to main content
Glama

start_vllm

Launch a local OpenAI-compatible vLLM server per model, bind it to 127.0.0.1, wait for health, and register it as a backend.

Instructions

Launch a local OpenAI-compatible server bound to 127.0.0.1 only (never 0.0.0.0). Each catalog model gets its own Docker container, port in 8000–8099, backend id (vllm- from the catalog id, e.g. vllm-qwen25-7b-instruct), and specialist. Tensor-parallels across every GPU only when the catalog fit needs more than one card; models that fit a single GPU stay at --tensor-parallel-size 1. Pass use_all_gpus=false to pin to one GPU even when the model is larger. Does not stop other orchestrator vLLM containers unless replace=true (restarts this model only). On intel-xpu, if intel/llm-scaler-vllm or intel/vllm:*xpu is local, starts that container (API published as 127.0.0.1:port:8000). Model must already be downloaded. Waits until GET /v1/models is healthy, upserts that backend + specialist, and stores a dummy loopback Bearer in gitignored GUI secrets if needed (never copy a key from vLLM). Cloud agents still cannot reach this server.

Input Schema

TableJSON Schema
NameRequiredDescriptionDefault
hostNoMust be 127.0.0.1 or localhost; any other value is rejected
portNo
imageNoLocal Docker image override, e.g. intel/vllm:0.17.0-xpu or intel/llm-scaler-vllm:0.21.0-b3
replaceNoIf true, stop the existing instance of this model before starting. Other running vLLM models are left alone.
runtimeNoForce Docker (Intel images) or host vllm. Default on intel-xpu is docker when a matching image exists.
model_idYesCatalog id or Hugging Face repo already downloaded
timeout_msNo
quantizationNo
use_all_gpusNoAllow using every GPU when the catalog fit needs tensor parallel. False: pin to one GPU. Models that already fit one card stay on one GPU even when this is true.
Behavior5/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

No annotations are provided, so the description fully carries the burden of behavioral disclosure. It covers binding restrictions, container isolation, tensor-parallel scaling, health checking, backend/specialist upsert, dummy Bearer storage ('never copy a key from vLLM'), and the fact that cloud agents cannot reach the server. This is exceptionally detailed for a tool without annotations.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness5/5

Is the description appropriately sized, front-loaded, and free of redundancy?

The description is long yet dense; every sentence introduces a distinct fact (binding, containerization, GPU scaling, replace semantics, intel-xpu special case, health wait, secrets handling). The most important scoping constraint ('bound to 127.0.0.1 only') is front-loaded at the very beginning.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness5/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

For a 9-parameter tool with no output schema or annotations, the description covers prerequisites, side effects, platform differences, security, and idempotency. It also describes the completion condition (waits until GET /v1/models is healthy) and the state changes performed, making the tool fully actionable.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters4/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema descriptions cover 67% of parameters, and the description adds meaning beyond them—e.g., clarifying the default tensor-parallel behavior, the intel-xpu container selection, the backend id derivation (vllm-<slug>), and the port range. It does not explain timeout_ms or quantization, but the schema constraints/enums partially cover those.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

The description states a specific verb ('Launch'), a resource ('local OpenAI-compatible server'), and adds precise constraints (bound to 127.0.0.1 only, port 8000–8099, per-model Docker container). It clearly differentiates from sibling tools like stop_vllm, remove_vllm, and vllm_status by focusing on starting a vLLM server.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines4/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

The description provides a prerequisite ('Model must already be downloaded'), explains when to use use_all_gpus=false, and gives platform-specific behavior on intel-xpu. It also clarifies the replace semantics relative to other running containers. However, it never explicitly names sibling alternatives such as download_local_model or vllm_status, leaving some comparison implicit.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

Install Server

Other Tools

Latest Blog Posts

MCP directory API

We provide all the information about MCP servers via our MCP API.

curl -X GET 'https://glama.ai/api/mcp/v1/servers/Unaware-Kerbin/agent-orchestrator'

If you have feedback or need assistance with the MCP directory API, please join our Discord server