You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Status update succeeds but list view keeps stale duplicate version active/isLatest #1819
Repro: publish the same version twice (e.g. publish 0.7.16 on 4 Nov, publish 0.7.16 again on 5 Nov after a metadata fix for a different server name+same name). Then publish a newer version (0.7.17). The list endpoint (GET /v0/servers?search=...) then returns BOTH the 4 Nov and 5 Nov rows for 0.7.16 side by side, the 5 Nov one still status: active, isLatest: true alongside the newer version. Running the publisher status call (mcp-publisher status --status deprecated <name> 0.7.16) twice returns 'Successfully updated status' both times ('deprecated -> deprecated'), but the list view still shows the 5 Nov row active+isLatest. The per-version endpoint GET /v0/servers/{name}/versions/{version} returns the deprecated 4 Nov row (statusChangedAt reflects the update, isLatest false), so the update only reached one duplicate. Duplicate-version entries seem to have no individually addressable id, and the stale twin keeps a misleading isLatest flag for crawlers. Expected: either reject republishing an identical version (forces a bump), or give duplicates individually addressable status updates so isLatest can be corrected.
We checked and found two different servers with version 0.7.16 under io.github.floortomproductions-droid:
agentx-x402-gateway — published November 4.
agentx-x402-services — published November 5.
Updating one doesn’t change the other. Both are now deprecated; services still has isLatest: true because 0.7.16 is its newest version.
Our tests rejected publishing the same full name/version twice, and listing and detail responses matched.
Does this explain what you saw? If both rows had the exact same full name and version, please share the request URL and response so we can investigate further.
floortomproductions-droid commented on Nov 12, 2026
Thanks @rdimitrov — yes, this explains what we saw, and to close the loop honestly:
The two 0.7.16s were two different names, not one name twice. When we filed on 7 Nov, the list showed a 0.7.16 row as active + isLatest: true right after two status --status deprecated calls that each returned "Successfully updated" — we read that as a stale duplicate under agentx-x402-gateway. In hindsight that row was agentx-x402-services/0.7.16, the other name's newest row, whose isLatest: true is by design regardless of status. We over-trusted the name match in the list and under-verified the identity — our misread, not a registry bug.
Current state, verified today on both the list and per-version endpoints: agentx-x402-gateway/0.7.16 → deprecated, isLatest: false; agentx-x402-services/0.7.16 → deprecated, isLatest: true; single row per name+version everywhere; and the only active record under either name is now agentx-x402-gateway @ 0.7.19. The confusing state has fully resolved.
One semantics question while you're here, since directories key off it: when the newest version of a name is deprecated, the list keeps isLatest: true on that row (ours does today — agentx-x402-services/0.7.16 shows deprecated + isLatest: true). Crawlers and directories we're listed in picked that flag up and showed "latest version" pointing at a deprecated row, which is what sent us down this path. Would you consider either a latestActive/status-aware flag, or a doc line making explicit that isLatest is status-independent so clients filter status themselves? Either answer works for us — we'd just rather build to the convention than around it.
If the above matches your read of both threads, happy for this to be closed as confirmed by-design behaviour.
Thanks for checking both endpoints and confirming the two records belonged to different full server names. That matches our findings.
Deprecating a version does not, by itself, remove its isLatest flag. A version can be both deprecated and the latest version of that server. version=latest and /versions/latest therefore do not guarantee an active result; directories should check _meta["io.modelcontextprotocol.registry/official"].status separately. If a directory wants to fall back to an older active version, it should inspect the server's version history and apply its own selection policy. Filtering out a deprecated latest result does not automatically select an older version.
We’re keeping these semantics rather than adding a latestActive flag. I’ve opened #1854 to make this explicit in the API and metadata documentation, including the OpenAPI descriptions. This wording is specific to deprecation; deleted versions have separate selection rules.
Closing as confirmed intended behavior, with no duplicate-version or stale-status bug reproduced. Thanks for helping identify the documentation gap.
Activity
rdimitrov commentedon Nov 12, 2026
We checked and found two different servers with version
0.7.16underio.github.floortomproductions-droid:agentx-x402-gateway— published November 4.agentx-x402-services— published November 5.Updating one doesn’t change the other. Both are now deprecated;
servicesstill hasisLatest: truebecause0.7.16is its newest version.Our tests rejected publishing the same full name/version twice, and listing and detail responses matched.
Does this explain what you saw? If both rows had the exact same full name and version, please share the request URL and response so we can investigate further.
floortomproductions-droid commentedon Nov 12, 2026
Thanks @rdimitrov — yes, this explains what we saw, and to close the loop honestly:
The two 0.7.16s were two different names, not one name twice. When we filed on 7 Nov, the list showed a
0.7.16row asactive+isLatest: trueright after twostatus --status deprecatedcalls that each returned "Successfully updated" — we read that as a stale duplicate underagentx-x402-gateway. In hindsight that row wasagentx-x402-services/0.7.16, the other name's newest row, whoseisLatest: trueis by design regardless of status. We over-trusted the name match in the list and under-verified the identity — our misread, not a registry bug.Current state, verified today on both the list and per-version endpoints:
agentx-x402-gateway/0.7.16→deprecated,isLatest: false;agentx-x402-services/0.7.16→deprecated,isLatest: true; single row per name+version everywhere; and the only active record under either name is nowagentx-x402-gateway @ 0.7.19. The confusing state has fully resolved.One semantics question while you're here, since directories key off it: when the newest version of a name is deprecated, the list keeps
isLatest: trueon that row (ours does today —agentx-x402-services/0.7.16showsdeprecated+isLatest: true). Crawlers and directories we're listed in picked that flag up and showed "latest version" pointing at a deprecated row, which is what sent us down this path. Would you consider either alatestActive/status-aware flag, or a doc line making explicit thatisLatestis status-independent so clients filterstatusthemselves? Either answer works for us — we'd just rather build to the convention than around it.If the above matches your read of both threads, happy for this to be closed as confirmed by-design behaviour.
rdimitrov commentedon Nov 12, 2026
Thanks for checking both endpoints and confirming the two records belonged to different full server names. That matches our findings.
Deprecating a version does not, by itself, remove its
isLatestflag. A version can be bothdeprecatedand the latest version of that server.version=latestand/versions/latesttherefore do not guarantee an active result; directories should check_meta["io.modelcontextprotocol.registry/official"].statusseparately. If a directory wants to fall back to an older active version, it should inspect the server's version history and apply its own selection policy. Filtering out a deprecated latest result does not automatically select an older version.We’re keeping these semantics rather than adding a
latestActiveflag. I’ve opened #1854 to make this explicit in the API and metadata documentation, including the OpenAPI descriptions. This wording is specific to deprecation; deleted versions have separate selection rules.Closing as confirmed intended behavior, with no duplicate-version or stale-status bug reproduced. Thanks for helping identify the documentation gap.
docs: clarify that deprecation does not remove the latest flag (#1854)