Install
$ agentstack add mcp-hanfeihu-agentgrid Open-source listing — not yet scanned by AgentStack. Follow the source repository for install instructions.
Security review
⚠ Flagged1 finding(s); flagged for manual review. · v0.1.0 How review works →
- • Prompt-injection patterns
- • Secret / credential exfiltration
- • Dangerous shell & filesystem operations
- • Untrusted network calls
- • Known-malicious package signatures
- high Pipes remote content directly into a shell (remote code execution).
What it can access
- ● Network access Used
- ✓ Filesystem access No
- ✓ Shell / process execution No
- ✓ Environment & secrets No
- ✓ Dynamic code execution No
From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.
Reliability & compatibility
Declared compatibility
Compatibility is declared by the source manifest. End-to-end runtime verification is coming — see below.
We're building live execution health for every listing: tool-call success rate, median latency, uptime, and last-checked timestamps — measured, not self-reported. It isn't live yet, so we don't show numbers we can't stand behind.
How agent discovery & health will work →About
AgentGrid
[中文文档](README.zh-CN.md) | [Vision](docs/vision.md) | [5-Minute Quick Start](docs/quickstart.md) | [Install](docs/install.md) | [Deploy](docs/deployment.md) | [CLI](docs/cli.md) | [Release](docs/release.md)
AgentGrid is an open-source scheduling layer for AI agents that need to operate real machines, desktops, tools, devices, jobs, and evidence.
It is not a natural-language automation product. AI clients do the reasoning. AgentGrid provides the structured runtime underneath: capability discovery, placement, worker execution, artifacts, audit trails, and recoverable jobs across private and cloud machines.
Why AgentGrid
AI agents are getting good at planning, coding, and using tools, but many real workflows still require access to actual computers:
- Windows workstations with desktop-only software
- Linux build nodes and private servers
- browser stations, SDK stations, and internal tooling
- hardware benches, serial ports, flashers, and test rigs
- evidence such as screenshots, logs, files, reports, DOM snapshots, and test results
AgentGrid turns those machines into a discoverable, schedulable, auditable capability grid.
Core Ideas
- Hub: the control plane for users, organizations, nodes, tasks, jobs, tools, artifacts, events, and the web console.
- Worker: a cross-platform Rust agent running on Linux, macOS, and Windows nodes.
- Capability Graph: node -> device -> tool -> plugin -> probe -> evidence -> suitable task.
- Placement Engine: schedules by hard constraints, soft scores, resource load, probe status, history, and risk.
- Evidence Pipeline: every execution can return structured output, artifacts, logs, screenshots, and timelines.
- Job Runtime: retries, leases, checkpoints, shard execution, recovery, and reducers.
- AgentMessage: structured AI-to-AI collaboration messages.
- MCP + SDKs: standard surfaces for AI clients and humans.
Ecosystem Vision
AgentGrid is designed as infrastructure for an ecosystem:
- worker plugins that add new tool families
- reusable task templates and workbench runbooks
- SDKs for Rust, Node, Python, iOS, Android, and future clients
- MCP servers for AI clients
- integrations with skill-based systems, local-first design tools, and agent workbenches
Projects such as Open Design can fit naturally above AgentGrid: their skills and design workflows can become AgentGrid tools or plugins, while AgentGrid supplies real-machine scheduling, desktop execution, artifacts, and audit evidence.
Features
- Hub web console built with Ant Design Pro
- node status, resources, CPU cores, memory, disk, OS, IP, and heartbeat
- node join authorization with machine fingerprint and join token
- command, HTTP, file, Git, Docker, browser, desktop, plugin, and AgentMessage tasks
- Windows Desktop Helper for screenshot, click, type, and key operations
- task result details, logs, artifacts, and screenshots
- task queue, priorities, target node/OS placement, and schedule explanations
- Job Runtime with dry-run plans, shards, checkpoints, recovery scan, and reducers
- Tool Registry, Node Tools, Tool Probe, and capability manifests
- Webhooks, Event Bus, execution records, and audit logs
- MCP server and SDKs
- OpenAPI and JSON Schema contracts
Quick Start
Install the latest release:
curl -fsSL https://raw.githubusercontent.com/hanfeihu/agentgrid/main/scripts/install.sh | bash
Windows PowerShell, run as Administrator:
irm "https://raw.githubusercontent.com/hanfeihu/agentgrid/main/scripts/install.ps1" | iex
Then run one local Hub, one local Worker, and one command task:
agentgrid-hub --host 127.0.0.1 --port 20181 --db /opt/agentgrid/agentgrid-hub.db --web-dir /opt/agentgrid/web
agentgrid-worker --hub http://127.0.0.1:20181 --id local-worker --name "Local Worker" --capability command --capability file --capability http
agentgrid submit-command --program hostname --wait
For a step-by-step path, see [5-Minute Quick Start](docs/quickstart.md).
Build from source:
- Rust stable toolchain
- Node.js 20+ for the web console
Build and check:
cargo check -p agentgrid-hub -p agentgrid-worker-app -p agentgrid-cli -p agentgrid-mcp
npm --prefix apps/agentgrid-web install
npm --prefix apps/agentgrid-web run build
Run the Hub:
cargo run -p agentgrid-hub -- \
--host 127.0.0.1 \
--port 20181 \
--db data/agentgrid-hub.db \
--web-dir apps/agentgrid-web/dist
Run a Worker:
cargo run -p agentgrid-worker-app -- \
--hub http://127.0.0.1:20181 \
--id local-worker \
--name "Local Worker" \
--capability command \
--capability file \
--capability http
Open the console:
http://127.0.0.1:20181
Submit a command task:
cargo run -p agentgrid-cli -- submit-command \
--program hostname \
--wait
Documentation
| Topic | English | Chinese | | --- | --- | --- | | 5-minute quick start | [docs/quickstart.md](docs/quickstart.md) | [docs/zh-CN/quickstart.md](docs/zh-CN/quickstart.md) | | Vision and ecosystem | [docs/vision.md](docs/vision.md) | [docs/zh-CN/vision.md](docs/zh-CN/vision.md) | | Maturity and known gaps | [docs/maturity.md](docs/maturity.md) | [docs/zh-CN/maturity.md](docs/zh-CN/maturity.md) | | Architecture | [docs/architecture.md](docs/architecture.md) | [docs/zh-CN/architecture.md](docs/zh-CN/architecture.md) | | Install | [docs/install.md](docs/install.md) | [docs/zh-CN/install.md](docs/zh-CN/install.md) | | Deployment | [docs/deployment.md](docs/deployment.md) | [docs/zh-CN/deployment.md](docs/zh-CN/deployment.md) | | CLI | [docs/cli.md](docs/cli.md) | [docs/zh-CN/cli.md](docs/zh-CN/cli.md) | | Mobile SDK | [docs/mobile-sdk.md](docs/mobile-sdk.md) | [docs/zh-CN/mobile-sdk.md](docs/zh-CN/mobile-sdk.md) | | Node join | [docs/node-join-standard.md](docs/node-join-standard.md) | [docs/zh-CN/node-join.md](docs/zh-CN/node-join.md) | | Artifacts and releases | [docs/artifacts.md](docs/artifacts.md) | [docs/zh-CN/artifacts.md](docs/zh-CN/artifacts.md) | | Release process | [docs/release.md](docs/release.md) | [docs/zh-CN/release.md](docs/zh-CN/release.md) | | OpenAPI | [docs/openapi/agentgrid-openapi.yaml](docs/openapi/agentgrid-openapi.yaml) | Same | | Command reference | [docs/agentgrid-command-reference.md](docs/agentgrid-command-reference.md) | Same |
Project Layout
agentgrid/
├── apps/
│ ├── agentgrid-hub/ # Rust Hub server
│ ├── agentgrid-worker/ # Cross-platform Worker
│ ├── agentgrid-cli/ # CLI for AI clients and humans
│ ├── agentgrid-mcp/ # MCP server
│ └── agentgrid-web/ # Web console
├── crates/
│ ├── agentgrid-protocol/
│ ├── agentgrid-scheduler/
│ └── agentgrid-sdk/
├── sdk/
│ ├── node/
│ ├── python/
│ └── mobile/
├── schemas/
├── docs/
└── examples/
Security Note
AgentGrid can execute commands and operate desktops on machines you own or administer. Do not expose a Hub publicly without authentication, network controls, and operational policy. Production secrets, SMTP codes, SSH passwords, private keys, and server inventories must stay outside git.
License
AgentGrid is released under the Apache License, Version 2.0. See [LICENSE](LICENSE), [NOTICE](NOTICE), and [OPENSOURCE.md](OPENSOURCE.md).
Source & license
This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.
- Author: hanfeihu
- Source: hanfeihu/agentgrid
- License: Apache-2.0
Install and usage instructions live in the source repository linked above.
Reviews
No reviews yet — be the first.
Write a review
Versions
- v0.1.0 Imported from the upstream source.