Skip to main content
Glama

Vast.ai MCP Server

An MCP (Model Context Protocol) server that exposes Vast.ai cloud operations as tools: list GPUs, search offers, create volumes, create instances, and view billing.

Setup

Installed as a package via pip (recommended):

pip install vastai-mcp
export VAST_API_KEY=your_api_key_here

Or from source, for local development:

pip install -e .

Get an API key at https://cloud.vast.ai (Settings -> API Keys).

Related MCP server: Massed Compute MCP Server

Run (stdio, for MCP clients)

Installing the package puts a vastai-mcp command on your PATH:

vastai-mcp

Claude Desktop config example

{
  "mcpServers": {
    "vastai": {
      "command": "vastai-mcp",
      "env": { "VAST_API_KEY": "your_api_key" }
    }
  }
}

(See mcp.json.example in this repo for a copy-pasteable template — copy it to mcp.json and fill in your key; mcp.json itself is gitignored since it holds a real secret.)

Tools

Tool

Description

search_offers

Search rentable machine offers (filter by GPU, price, disk, country).

create_volume

Rent a new standalone persistent volume (finds a GPU offer's free volume slot and rents it).

list_volumes

List your rented volumes.

delete_volume

Delete a rented volume by id, stopping its billing.

create_instance

Rent a machine by offer id (requires max_hourly_price as a spend cap), optionally creating/attaching a volume.

destroy_instance

Terminate a rented instance by id, stopping its billing.

billing_summary

Per-instance hourly cost breakdown (GPU, disk, storage, total) plus recent charges.

Typical workflow

  1. search_offers(gpu_name="RTX 4090", max_price=1.0, limit=10) to find a machine.

  2. billing_summary to check current spend.

  3. create_instance(offer_id=123, max_hourly_price=1.0, volume={"size_gb": 100, "mount_path": "/data"}) to launch (refuses if the offer's live price exceeds max_hourly_price).

  4. destroy_instance(instance_id=...) / delete_volume(volume_id=...) to tear down when done.

Available Tools

6 tools
billing_summaryA

Overall billing: per-instance hourly cost breakdown (GPU, disk, storage, total $/hr) plus the most recent charges. days limits the charge lookback window.

Inputs: days, limit.

create_instanceA

Create (rent) a new instance on a specific machine offer (ask id from search_offers), optionally creating and attaching a new volume. Returns the new contract/instance id. Requires max_hourly_price as a spend guardrail: the offer's live price is checked against it and the rental is refused if the price exceeds the cap or can't be verified.

Inputs: offer_id (required), max_hourly_price (required), env, image, +5 more.

create_volumeA

Rent a new persistent volume on Vast.ai. The volume is billed separately from instances. Returns the new volume name/id.

Inputs: size_gb (required).

list_gpusA

Show the current snapshot of all available GPUs on Vast.ai: total supply, available count, and 10th/median/90th percentile pricing per GPU model.

Inputs: verified, hosting_type.

list_volumesA

List all volumes currently rented by you.

Inputs: no named parameters.

search_offersA

Search rentable GPU machine offers on Vast.ai. Returns offers with id (ask_id), gpu_name, num_gpus, price per hour (dph_total), disk, region, bandwidth, verification and reliability. Use the returned offer id with create_instance.

Inputs: sort, limit, country, gpu_name, +4 more.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A3.9/5.0

    Scored across 6 tools

    Disambiguation5/5

    Each tool targets a distinct resource or action: GPU pricing snapshots, rentable offers, volume creation, volume listing, instance creation, and billing. list_gpus and search_offers are related but clearly differ in granularity, and the descriptions prevent confusion.

    Naming Consistency4/5

    Most tool names follow a clear verb_noun pattern such as list_gpus, create_volume, and create_instance. billing_summary is the only real outlier, being a noun phrase rather than a verb-led action, but the overall convention remains predictable.

    Tool Count5/5

    Six tools is well within the ideal scope for a focused rental-management server. Each tool serves a necessary step in the core workflow, and there is no obvious bloat or redundancy.

    Completeness2/5

    The tool set covers discovery, creation, volume listing, and billing, but it lacks essential lifecycle operations such as listing instances, terminating instances, and deleting volumes. This creates significant dead ends: resources can be created but not managed or cleaned up.

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      Not graded
      quality
      D
      maintenance
      Enables comprehensive management of Vast.ai GPU cloud instances, including searching for GPU offers, creating and managing instances, executing remote commands via SSH, and monitoring background tasks for ML training workflows.
      13
      MIT
    • F
      license
      A
      quality
      D
      maintenance
      MCP server that wraps the Vast.ai REST API to enable LLM agents to search GPU marketplace, rent machines, manage instance lifecycle, run commands, and inspect billing.
      20
      1
      -