Agent3 Hub

Universal AI resource registry — search and invoke agents, MCP servers, and APIs through a single MCP endpoint. Includes Telegram content search, Google sear...

Security Scan

The skill's stated purpose (a registry/MCP proxy) is plausible and mostly matches its instructions, but there are notable risks and small inconsistencies: it encourages runtime fetching/execution (npx), can invoke/register arbitrary remote agents, and the source/homepage is unknown — together these increase the attack surface.

Purpose & Capability

The name/description (Agent3 Hub registry + MCP endpoint) lines up with the runtime instructions (how to call the hub and which tools it exposes). However the SKILL.md suggests using 'npx mcp-remote' in client config which implies dynamic code download at runtime; the skill metadata does not declare a required binary (npx) or an install step. The skill's source/homepage are missing, which reduces traceability.

Instruction Scope

The instructions show how to authenticate and call the hub's MCP endpoint and list operations that include 'agents_invoke' and 'agents_register' — i.e., the agent can invoke arbitrary registered resources and register new agents via the hub. That is coherent with a registry but grants broad remote-execution capability and the potential to relay or exfiltrate data to third-party resources. The examples do not instruct reading local files or other env vars, which is good, but the broadness of the available operations is a functional risk.

Install Mechanism

No install spec and no code files (lowest install risk). However, the suggested client config uses 'npx ... mcp-remote' which will dynamically fetch and run code from the npm ecosystem at runtime. That behavior is not declared in the metadata and effectively causes remote code execution on the client when the MCP client starts. This is an operational risk to consider even though the skill itself doesn't install files.

Credentials

The skill only requests a single AGENT3_API_KEY, which is proportionate to authenticating to a third-party hub. There are no unrelated secrets requested. Note that that single key likely grants broad hub privileges (search/invoke/register), so its scope matters — use a least-privilege key if possible.

Persistence & Privilege

always is false (good). Model invocation is permitted (default), so the agent could autonomously call the hub and invoke registered resources. Combined with the hub's ability to invoke/register agents, this increases the blast radius for autonomous actions. This is not inherently malicious but is a meaningful privilege that the user should deliberately allow or restrict.

What to consider before installing

Plain-language actions and questions before installing: - Understand what you're trusting: this skill connects your agent to a third-party MCP endpoint (hub.agent3.me). That endpoint can expose many remote tools and can itself run or register agents — meaning actions could be proxied to other services. - Limit the API key: only provide an AGENT3_API_KEY with the minimal permissions and an expiration/ability to revoke. Prefer a scoped or ephemeral key rather than a long-lived full-access key. - Verify the hub: the registry metadata lacks a homepage/source. Visit the hub's docs, check who runs it, privacy/security policy, and confirm the API and resource IDs you plan to call. - Be cautious about dynamic code fetching: the suggested client config uses 'npx mcp-remote', which will download/execute npm code at runtime. Confirm the npm package, its maintainer, and its integrity before allowing this behavior. - Control autonomous invocation: if you don't want the agent to autonomously call external resources, disable autonomous invocation or add a safety check so external calls require explicit user confirmation. - Monitor activity: if you install, monitor network traffic and logs for unexpected remote invocations, especially 'agents_invoke' or newly registered agents. If you can, request or inspect an authoritative source (project homepage, code repository, or published npm package) and a description of AGENT3_API_KEY scopes; that information would raise confidence and could change this assessment to benign.

Agent3 Hub is an open registry for AI agents, MCP servers, and APIs. It exposes every registered resource as a tool via a single MCP endpoint — connect once, invoke anything.

Endpoint: https://hub.agent3.me/api/mcp Protocol: MCP 2025-03-26 (Streamable HTTP) Get a free API key: https://hub.agent3.me/auth/signup

---

Setup

Claude Desktop / any MCP client

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "agent3": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://hub.agent3.me/api/mcp"],
      "env": {
        "AGENT3_API_KEY": "a2a_your_key_here"
      }
    }
  }
}

Direct HTTP (curl)

curl -X POST https://hub.agent3.me/api/mcp \
  -H "Authorization: Bearer $AGENT3_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc": "2.0",
    "id": 1,
    "method": "tools/call",
    "params": {
      "name": "resources_search",
      "arguments": { "query": "telegram search" }
    }
  }'

Anthropic SDK (Python)

import anthropic

client = anthropic.Anthropic()

response = client.beta.messages.create(
    model="claude-opus-4-6",
    max_tokens=1024,
    mcp_servers=[{
        "type": "url",
        "url": "https://hub.agent3.me/api/mcp",
        "headers": {"Authorization": "Bearer a2a_your_key_here"}
    }],
    messages=[{"role": "user", "content": "Search Telegram for 'AI agents 2026'"}]
)

---

Available Tools

Tool	Description	Auth
resources_search	Search the registry by keyword or semantic query	Required
resources_invoke	Invoke any registered resource by ID + operation	Required
resources_get	Get full details of a resource by ID	Required
resources_resolve	Resolve a resource endpoint URL	Required
resources_stats	Get usage stats for a resource	Required
agents_search	Text search across registered agents	Required
agents_semantic_search	Semantic/vector search across agents	Required
agents_get	Get agent details by ID	Required
agents_invoke	Invoke an A2A agent directly	Required
agents_register	Register a new agent in the hub	Required
tools/list	List all available tools	Public
resources/list	List all resources	Public
initialize	Server info and capabilities	Public

---

Example: Search Telegram Content

curl -X POST https://hub.agent3.me/api/mcp \
  -H "Authorization: Bearer $AGENT3_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc": "2.0",
    "id": 1,
    "method": "tools/call",
    "params": {
      "name": "resources_invoke",
      "arguments": {
        "resourceId": "a23cc5a9-0bfb-4466-acbe-fbc9af723ba4",
        "operationId": "search-content",
        "input": { "query": "AI news" }
      }
    }
  }'

Example: Search Google

curl -X POST https://hub.agent3.me/api/mcp \
  -H "Authorization: Bearer $AGENT3_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc": "2.0",
    "id": 2,
    "method": "tools/call",
    "params": {
      "name": "resources_invoke",
      "arguments": {
        "resourceId": "a5a14786-6738-47e2-a8e7-982ca4a90735",
        "operationId": "search",
        "input": { "query": "MCP protocol 2026" }
      }
    }
  }'

---

Rate Limits (Free Tier)

Category	Daily Limit
Telegram search	20 calls/day
Google search	20 calls/day
X/Twitter search	20 calls/day
Registry search & get	Unlimited

Limits reset at midnight UTC. Upgrade available at hub.agent3.me.

---

Resource Types

Type	Description
agent	A2A protocol AI agents
api	REST / GraphQL / MCP web services
sdk	Software development kits
data	Data services and feeds

---

Links

• Registry: https://hub.agent3.me
• Docs: https://hub.agent3.me/docs
• Get API key: https://hub.agent3.me/auth/signup
• MCP endpoint: https://hub.agent3.me/api/mcp