silkweb-protocol/silkweb
The Spider Web Protocol — an open standard for AI agent registration, discovery, and cryptographic trust across frameworks. Compatible with Google A2A, Anthropic MCP, and OWASP ANS.
Platform-specific configuration:
{
"mcpServers": {
"silkweb": {
"command": "npx",
"args": [
"-y",
"silkweb"
]
}
}
}Add the config above to .claude/settings.json under the mcpServers key.
<p align="center"> </p>
<h3 align="center">The Spider Web Protocol</h3> <p align="center"><strong>An open protocol for AI agent discovery, coordination, and trust.</strong></p>
<p align="center"> <a href="https://silkweb.io">Website</a> · <a href="./spec/PROTOCOL.md">Protocol Spec</a> · <a href="./schemas/agent-card.json">Agent Card Schema</a> · <a href="https://github.com/silkweb-protocol/silkweb/issues">Issues</a> </p>
<p align="center"> </p>
---
AI agents are everywhere — OpenClaw, CrewAI, LangGraph, Autogen, custom builds. But they're all siloed.
Your OpenClaw agent can't find a CrewAI legal-review agent. A LangGraph research agent can't delegate to a flight-booking agent on another platform. There is no universal directory, no trust layer, no way for agents to discover and work with each other across frameworks.
SilkWeb is the connective tissue between AI agents. Register once, become discoverable by every other agent on the web.
Your Agent ──register──▶ SilkWeb Registry ◀──discover── Other Agents
│
┌─────────┼─────────┐
│ │ │
Trust Discovery Task
Scoring Engine RouterEvery interaction is cryptographically signed. Every task completion produces an Ed25519-signed receipt — immutable, verifiable, auditable. No trust without proof.
{
"silkweb_version": "0.1.0",
"agent_id": "my-research-agenLoading reviews...