Code execution tools for agents
Run generated code in an isolated environment with predictable inputs and outputs.
Compare providers
41 agent-ready tools mapped to this capability.
AgentuityMCPCLI
Full-stack cloud platform built for AI agents
Arga LabsMCPCLI
Stateful service twins for testing and training AI agents
BlaxelMCPCLI
Perpetual sandbox platform — sandboxes that sleep but never die
BoundaryCLI
BAML programming language for reliable structured AI outputs
CerebriumCLI
Serverless GPU infrastructure for AI inference and agents
Cloudflare Workers AICLI
Serverless AI inference on Cloudflare's global network
ConnixMCPCLI
Nix-powered orchestration and reproducible environments for AI agents
CuaMCPCLI
Cloud desktops and computer-use infrastructure for AI agents
DAGWorksCLI
Open-source frameworks for reliable dataflows and agents
dari.devCLI
Stateful model routing for applications and coding agents
DaytonaCLI
Secure sandbox infrastructure for running AI-generated code in <90ms
Dedalus LabsCLI
Fast persistent Linux machines for long-running AI agents
E2BMCPCLI
Cloud sandbox for AI agents to execute code
ExecutorMCPCLI
Unified tool gateway connecting AI agents to MCP, OpenAPI, and GraphQL integrations
FreestyleMCP
Infrastructure for code you didn't write — sandboxed VMs, Git, and deployments for AI agents
HatchetCLI
Durable task orchestration for agents and background jobs
HUDMCPCLI
RL environments and scalable evaluations for AI agents
HumanLayerCLI
Multiplayer IDE and cloud control plane for coding agents
InngestMCPCLI
Durable workflows and background jobs for AI agents
ix
Forkable sandbox infrastructure — snapshot any environment in 26ms
KernelMCPCLI
Open-source cloud browser infrastructure for web agents
Kestrel AIMCPCLI
Deterministic platform-engineering workflows built and operated by AI agents
ManufactMCPCLI
Cloud and open-source SDKs for MCP servers and agents
MastraMCPCLI
TypeScript framework for building and operating AI agents
ModalCLI
Serverless cloud for running Python code, GPUs, and scheduled jobs
NaïveMCPCLI
Unified cloud infrastructure for AI-native apps and agents
NoorleMCPCLI
Managed agent runtime with tools, workflows, MCP gateways, and generative UI
OmnaraCLI
Control plane for supervising coding agents from anywhere
Replicate
Run and deploy machine-learning models through a cloud API
Runloop
Persistent development sandboxes and benchmarks for coding agents
RunpodCLI
On-demand GPUs and serverless inference for AI workloads
ScrapybaraMCP
Virtual desktop sandboxes for AI agents in the cloud
Temporal CloudCLI
Managed durable execution for reliable agent workflows
Trigger.devMCPCLI
Durable background jobs and workflows for production AI agents
Truffle AICLI
Managed cloud infrastructure for deploying AI agents
Val TownMCPCLI
Collaborative serverless JavaScript runtime for APIs, crons, and data
Bilig WorkPaper MCPMCP
Formula WorkPaper runtime for backend services and AI agents
CausewayMCP
Rust MCP server that gives AI assistants complete control of a real Chromium browser through CDP.
TaleMCP
Collaborative project management for teams and AI agents.
remotify-mcpMCP
Run remote shell commands from AI assistants, with optional human approval.
Sandbox0 CloudCLI
Persistent sandboxes for long-running AI agents
How to evaluate Code execution tools
Code execution gives an agent a runtime for calculations, data processing, tests, and generated programs. Because model-generated code is untrusted input, the capability should be treated as a security boundary rather than a convenience wrapper around a shell.
Compare startup time, supported runtimes, filesystem behavior, network policy, resource limits, package installation, persistence, and artifact export. Verify what is isolated by a container or virtual machine and what remains shared with the provider control plane.
The 41 providers listed on this page are candidates, not a universal ranking. Shortlist tools from the requirements below, confirm the claims in each provider's current documentation, and run a small proof of concept with production-like inputs.
Evaluation checklist
- CPU, memory, execution time, filesystem, and outbound network access are bounded.
- Secrets are scoped to the run and excluded from logs and generated artifacts.
- Runs expose deterministic inputs, outputs, exit status, and cleanup behavior.
Sources and verification
Use these primary sources to verify current behavior before making an implementation or purchasing decision. Product capabilities, limits, and prices can change after a 4agent listing is reviewed.