agentguard — policy proxy for agent tool calls
warningsagentwares-agentguard.vercel.app
Manage hosted agentguard proxies: caps, dry-run, loop breaker, run reports, audit export.
https://agentwares-agentguard.vercel.app/api/mcpWebsiteNo authMCP 2025-11-25v0.2.1last checked 22 Oct 2026spoke MCP on 100% of 88 checks (30 days)initialize in 218 msLast checked 22 Oct 2026
- ✓Answered the MCP handshake (MCP 2025-11-25)
- ✓Auth detected: No auth
- ✓Lists 7 tools
- !1 warning
- ✓Spoke MCP on 88 of 88 checks (30 days)
Will my client connect?
Decided from what the probe found at the endpoint, and from what each client's own documentation says it can complete. Every row carries the day that documentation was read.
✓Claude CodeConnects
The endpoint answers an anonymous handshake, so no credential is needed.
claude mcp add --transport http agentguard-policy-proxy-for-agent-tool-calls https://agentwares-agentguard.vercel.app/api/mcpClaude Code docs · read on 2026-08-30
✓Claude (claude.ai & Desktop)Connects
The endpoint answers an anonymous handshake, so no credential is needed.
Settings → Connectors → Add custom connector, then paste https://agentwares-agentguard.vercel.app/api/mcp.
Claude (claude.ai & Desktop) docs · read on 2026-08-30
✓ChatGPTConnects
The endpoint answers an anonymous handshake, so no credential is needed.
Settings → Apps → Advanced settings → Developer mode, then create a connector pointing at https://agentwares-agentguard.vercel.app/api/mcp.
ChatGPT docs · read on 2026-08-30
✓CursorConnects
The endpoint answers an anonymous handshake, so no credential is needed.
{
"mcpServers": {
"agentguard-policy-proxy-for-agent-tool-calls": {
"url": "https://agentwares-agentguard.vercel.app/api/mcp"
}
}
}Cursor docs · read on 2026-08-30
✓VS Code (Copilot)Connects
The endpoint answers an anonymous handshake, so no credential is needed.
{
"servers": {
"agentguard-policy-proxy-for-agent-tool-calls": {
"type": "http",
"url": "https://agentwares-agentguard.vercel.app/api/mcp"
}
}
}VS Code (Copilot) docs · read on 2026-08-30
How to authenticate
Checked against the endpoint by our probe on 22 Oct 2026. These are the capabilities the server advertises, not a prediction about any particular client.
| Scheme | No auth |
|---|---|
| Protected-resource metadata | ✕Not found Without it a client has to guess where the authorization server lives. |
| Anonymous access | Handshake succeeds; 7 tools listable |
| Credentials demanded in prose | ✕okgate_create_proxy, okgate_set_policy, okgate_get_run, okgate_export_audit, okgate_recommend_allowlist These tools are listed anonymously and their own descriptions ask for an account or key, but no challenge is issued — so a client has nothing to act on and the refusal arrives as a tool error. |
What the probe found
5 of the 7 tools listed anonymously say in their own descriptions that they need an account or a key (okgate_create_proxy, okgate_set_policy, okgate_get_run, okgate_export_audit, okgate_recommend_allowlist), but the endpoint issued no WWW-Authenticate challenge. The requirement is stated only in prose, so a call reaches the tool and comes back as a tool error instead of an authorization step.
Tools
Contract history
Instructions rewritten: okgate_get_pricing. The description an agent follows changed; the callable contract may not have.
Instructions rewritten: okgate_get_pricing. The description an agent follows changed; the callable contract may not have.
Breaking: agentguard_create_proxy, agentguard_export_audit, agentguard_get_pricing, agentguard_get_run, agentguard_recommend_allowlist, agentguard_set_policy, agentguard_spend_report
Instructions rewritten: agentguard_get_pricing. The description an agent follows changed; the callable contract may not have.
Watch this server
Save its public contract as a baseline and manage breaking-change email alerts.
Set up a watchRoute it through the gateway
Put this endpoint behind one find_tools / call_tool pair with the rest of your servers. Part of Pro.
Add to my gatewayOwn this server?
Sign in to claim this listing by proving control of its host.
Status badge
Status for a README, from the last probe — cached an hour, so up to seven hours behind the endpoint. Links back to this page.
[](https://mcpi.app/servers/agentguard-policy-proxy-for-agent-tool-calls)