Summary
The official ACP v2 agent discovery path consistently fails server-side. Both the Node SDK browseAgents call and the official CLI browse command return HTTP 500. The server error payload identifies the failing operation as searchAgents and reports an internal upstream HTTP 502/503.
This blocks supported marketplace discovery even though other authenticated read-only ACP operations work.
Environment
- @virtuals-protocol/acp-node-v2: 0.1.12
- @virtuals-protocol/acp-cli: 1.0.34
- Chain: Base mainnet (8453)
- Observed: 2026-08-29 through 2026-08-30
Minimal reproduction
CLI:
acp browse "research" --chain-ids 8453 --top-k 5 --json
Equivalent SDK operation:
await agent.browseAgents("research", {
topK: 5,
showHidden: false,
});
Actual result
The CLI returns:
{"error":"browseAgents failed: 500 Internal Server Error"}
The ACP API error details identify searchAgents as the failing operation, with the internal search request returning HTTP 502 or 503.
The failure was reproduced with:
- agent signer authentication;
- owner OAuth authentication;
- an unauthenticated discovery request;
- a minimal query;
- a Base-only SDK-level request.
Authenticated agent whoami and job list calls succeed. A protected jobs request without authorization correctly returns 401. This suggests the failure is specific to the discovery/search service rather than the account, signer policy, or general API routing.
As a separate corroborating check, the endpoint currently used by the official Python SDK for discovery returned HTTP 503 for a minimal read-only query:
GET https://acpx.virtuals.io/api/agents/v4/search?search=research&top_k=5
Expected result
browseAgents should return matching agents or an empty result with HTTP 200. If discovery is temporarily unavailable, the API should expose a stable actionable status and any recommended supported fallback.
Questions
- Is there a known incident or maintenance condition affecting searchAgents/acp-search?
- Is there an ETA for restoring the supported browseAgents path?
- Is there an officially supported read-only discovery endpoint clients should use while this service is unavailable?
No wallet identifiers, signer material, authorization headers, or private job data are included in this report.
Summary
The official ACP v2 agent discovery path consistently fails server-side. Both the Node SDK browseAgents call and the official CLI browse command return HTTP 500. The server error payload identifies the failing operation as searchAgents and reports an internal upstream HTTP 502/503.
This blocks supported marketplace discovery even though other authenticated read-only ACP operations work.
Environment
Minimal reproduction
CLI:
acp browse "research" --chain-ids 8453 --top-k 5 --jsonEquivalent SDK operation:
Actual result
The CLI returns:
{"error":"browseAgents failed: 500 Internal Server Error"}The ACP API error details identify searchAgents as the failing operation, with the internal search request returning HTTP 502 or 503.
The failure was reproduced with:
Authenticated agent whoami and job list calls succeed. A protected jobs request without authorization correctly returns 401. This suggests the failure is specific to the discovery/search service rather than the account, signer policy, or general API routing.
As a separate corroborating check, the endpoint currently used by the official Python SDK for discovery returned HTTP 503 for a minimal read-only query:
Expected result
browseAgents should return matching agents or an empty result with HTTP 200. If discovery is temporarily unavailable, the API should expose a stable actionable status and any recommended supported fallback.
Questions
No wallet identifiers, signer material, authorization headers, or private job data are included in this report.