Skip to main content
AgentKey supports 22 agents across CLIs, IDEs, chat clients, and desktop apps. The right install method depends on which agent you use: a one-line shell command handles most CLI and IDE clients automatically, a simple prompt instructs chat-based agents to pull down the skill from ClawHub, and Claude Desktop needs a short two-step desktop config. All three paths end in the same place — your agent authenticated with a master key and ready to call ~1,800 tools.
The fastest path is the console get-started flow. Pick your agent and the console gives you the exact command or prompt for your OS, pre-filled with your key. The full list of supported agents is on Supported Agents.

​
One-line install (CLI & IDE)

Running a single command is all it takes for CLI and IDE clients. The installer detects your agent, registers the MCP server, and writes your API key automatically — no manual config editing required. 
curl -fsSL https://agentkey.app/install.sh | bash
This method works for: Claude Code, Cursor, Windsurf, Gemini CLI, OpenCode, Codex, Cursor CLI, Warp, Kimi CLI, Qwen Code, Kiro CLI, Amp, Crush, iFlow CLI.
Both commands download a script and execute it immediately. If you prefer to review the script before running it, download it first and inspect it:
curl -fsSL https://agentkey.app/install.sh -o install.sh
less install.sh   # read it over, then:
bash install.sh
This is good practice any time you pipe a remote script directly into your shell.

​
Prompt install (chat agents)

Chat-based agents don’t have a shell to run commands in, so you install AgentKey by pasting a prompt. The agent fetches the skill from ClawHub, reads its metadata, and walks you through the remaining setup steps — asking before making any changes beyond the skill itself. Paste this prompt exactly into your agent’s chat box: 
Install the skill chainbase/agentkey from ClawHub: https://clawhub.ai/chainbase/agentkey
Scope the work to this skill only. After install, read the skill's metadata and
help me finish setup based only on what you can verify from that page — don't
invent missing requirements. Ask before making any broader environment changes.
This method works for: OpenClaw, Hermes, Antigravity, Trae, Qoder, WorkBuddy, Cowork.

chainbase/agentkey on ClawHub

The official AgentKey skill page on ClawHub — the source the prompt above points your agent to for install details and metadata.

​
Claude Desktop (desktop config)

Claude Desktop requires two pieces: the MCP server registered on your machine, then the AgentKey skill file imported into the app.
1

Configure the MCP server

Run the one-line installer to detect Claude Desktop and register the MCP server in its config:
curl -fsSL https://agentkey.app/install.sh | bash
2

Download the AgentKey skill file

Download agentkey.skill from GitHub Releases. The /latest/ URL always resolves to the newest release, so you don’t need to track version numbers manually.
3

Import the skill into Claude Desktop

Drag the agentkey.skill file into the Claude Desktop chat box, or double-click it while Claude Desktop is running. Claude confirms with a “Skill imported” message when it’s ready.

​
Finish setup

After installing, you need to connect your master key so AgentKey can authenticate your requests and track usage against your credit balance. If you used the one-line installer, this step happens automatically. For prompt and desktop installs, follow the in-agent setup prompts.
1

Create a key

Sign in to the AgentKey console and generate a new key. Your key starts with ak_.
2

Add credits

Top up your balance once. Credits are shared across every tool and service you access through AgentKey — you don’t need separate billing for each one.
3

Provide the key to your agent

The one-line installer writes the key to your agent’s config automatically. For prompt and desktop installs, supply the key when your agent asks during its setup flow. Never paste your key directly into a chat message.
Your master key is a secret tied to your credit balance. Never commit it to source control or share it in a public forum. For more on keeping your key safe, see Authentication.

​
Verify it works

Once setup is complete, try a task that requires a live external tool — for example, ask your agent:
Search X/Twitter for what people are saying about AgentKey today.
Your agent should discover the right tool, look up its parameters, and execute the call. If it does, AgentKey is working. For a deeper look at how that discovery-to-execution flow works, see Discover → Describe → Execute.

​
Next steps

Supported Agents

All 22 supported agents and which install method each one uses.

Browse Capabilities

Explore the ~1,800 tools your agent can now reach through AgentKey.