Summary: On two freshly provisioned Apigee evaluation organizations (asia-east1 and us-central1), every MCP Discovery Proxy deployment stays in Provisioning for 38–55 minutes and then fails with MCP just-in-time provisioning has not completed. The runtime returns 503 protocol.http.NoResolvedHost for ORG.mcp.apigee.internal the whole time. The documentation states MCP is available for evaluation orgs, and known issue 496552286 (regional capacity) does not explain a failure in us-central1. I would like to know whether MCP just-in-time provisioning is actually supported on evaluation orgs / non-Comprehensive environments.
Environment
| Item | Org A | Org B |
|---|---|---|
| Project / org ID | apigee-eval-508708 |
apigee-eval-us |
| Org type | Evaluation (60-day trial) | Evaluation (60-day trial) |
| Runtime region | asia-east1 | us-central1 |
| Instance | eval-instance, runtime version 1-18-0-apigee-4 |
eval-instance, runtime version 1-18-0-apigee-4 |
| Environment / group | eval / eval-group (no environment type shown in console) |
eval / eval-group |
| Env group hostname | 34.111.166.123.nip.io (external LB, regular proxies serve 200) |
apigee-eval-us.dns-replaceme.example.com (default) |
| API hub | Provisioned in same region, runtime project auto-attached (Project associations tab shows the project) | same |
| OpenAPI spec | 3.0.3, 2 operations (getSampleUser, getCallerIp), servers.url == env group hostname |
same spec, servers.url == env group hostname |
| Proxy | mock-mcp from the MCP Discovery Proxy template; target endpoint https://ORG.mcp.apigee.internal/mcp |
same |
Attempts (all times UTC, 2026-09-17/18)
| # | Org | Deployed at | Deploy method | Service account | Result |
|---|---|---|---|---|---|
| 1 | A | 09-17 09:21 | Create wizard (create + deploy) | none | Failed after ~47 min |
| 2 | A | 09-17 23:51 | Undeploy → redeploy (wizard) | none | Failed after ~55 min |
| 3 | A | 09-18 02:13 | Undeploy → Deploy pane | mcp-deployer@… (SA specified) |
Failed after ~52 min |
| 4 | B | 09-18 04:13 | Create wizard (create + deploy) | none | Failed after ~38 min |
Error shown on the deployment status icon (identical wording in all four cases):
Status: no MCP configuration status has been reported for deployment
"organizations/<org>/apis/mock-mcp/revisions/1"; MCP just-in-time provisioning has not completed
Runtime response during and after each attempt (org A, ~55 probes, one per minute):
POST https://34.111.166.123.nip.io/mcp {"jsonrpc":"2.0","id":1,"method":"tools/list","params":{}}
HTTP 503
{"fault":{"faultstring":"Unable to resolve host apigee-eval-508708.mcp.apigee.internal",
"detail":{"errorcode":"protocol.http.NoResolvedHost","reason":"TARGET_CONNECT_HOST_NOT_REACHABLE"}}}
Cloud Audit Logs in the customer project show DeployApiProxy accepted, then one ReportInstanceStatus error (FAILED_PRECONDITION: request is missing required specs, resource environments/eval/deployedConfig) about 60–70 s after every deployment — including successful deployments of ordinary reverse proxies (hello-world, mock-target), so this appears unrelated. No other errors are visible from the customer project; the MCP provisioning presumably happens in the tenant project.
What has been ruled out
- OpenAPI version (3.0.3, within 3.0.0–3.0.3) and
servers.url/ env group hostname match. - API hub not provisioned / runtime project not attached (both verified in console).
- Environment unusable (regular reverse proxy
mock-targetdeploys and serves 200 in the same environment). - Runtime version too old (both instances run 1-18-0-apigee-4, newer than the 1-17-0-apigee-6 release that accompanied MCP GA).
- Missing deploy-time service account (attempt 3 specified one; no change).
- Regional capacity (known issue 496552286): us-central1 is not in the published list and fails identically.
Questions
- Is MCP just-in-time provisioning supported on evaluation organizations? The overview and quickstart say MCP is available for evaluation orgs, but the quickstart’s deploy step also says the environment “must be a Comprehensive environment”, and evaluation environments have no environment type.
- If it is supported, is there a way to see the provisioning error from the customer side (an operation, a log, or a status field)? Nothing is visible in Cloud Audit Logs beyond the accepted
DeployApiProxy. - If it is not supported, could the documentation be updated so evaluators do not spend time on it?
Happy to provide the org IDs above, exact timestamps, or run a debug session if useful.
