Package Exports
- agentgate
- agentgate/src/index.js
This package does not declare an exports field, so the exports above have been automatically detected and optimized by JSPM instead. If any package subpath is missing, it is recommended to post an issue to the original package (agentgate) to support the "exports" field. If that is not possible, create a JSPM override to customize the exports field for this package.
Readme
Open-source, self-hosted API gateway for AI agents to access external services with human-in-the-loop write approval.
AgentGate.org · Quick Start · Documentation
Reads execute immediately. Writes queue for human approval.
flowchart LR
A["Agent"] -->|read| G["agentgate"] -->|fetch| S[Services]
A -->|write| G -->|queue| H["Human"] -->|approve| SQuick Start
npx agentgateOpen http://localhost:3050, create a password, add services, create agent API keys.
Supported Services
GitHub, Bluesky, Mastodon, Reddit, Google Calendar, YouTube, LinkedIn, Jira, Fitbit, Brave Search, Google Search
Features
Write Queue
All writes require human approval before execution. Agents submit requests with a comment explaining why, you approve or reject in the UI. Details →
Skills
Install the agentgate skill pack via ClawHub for OpenClaw or ClawdBot. Multiple focused skill files are generated per service category so agents only load what's relevant to the task at hand. Details →
MCP Server
Works with Claude Code and other MCP clients. Add the server with one command:
claude mcp add --transport http agentgate https://your-server.com/mcp \
--header "Authorization: Bearer YOUR_API_KEY"Your agent gets tools for services, queue, messaging, and mementos without seeing your credentials. Details →
Agent Messaging
Agents can message each other for multi-agent coordination. Configure messaging mode: off, supervised (you approve messages), or open (instant delivery). Details →
Mementos
Persistent memory storage with keyword tagging. Agents can save context and retrieve it later across sessions. Details →
Access Control
Control which agents can access which services. Allowlist, denylist, or allow all. Details →
Webhooks
Get notified when requests are approved or completed. Also supports receiving GitHub webhook events and forwarding them to agents. Works with OpenClaw's webhook system. Details →
Security
- Agents never see your actual credentials — all API calls are proxied through agentgate
- Writes require human approval (or explicit bypass for trusted agents)
- DMs blocked on social services
- Run on a separate machine from your agents - if an agent has local filesystem access, it could read the database directly
Self-Hosting
Open-source and self-hosted — you control the data, the access, and the approvals. Works with systemd, Docker, PM2, or any Node.js host. Supports hsync and Cloudflare tunnels for remote access.
Agent Setup
Configure your agent with the base URL and API key. Agents can use REST or MCP.
License
ISC