Skip to content

This release adds 2 notable features for engineering teams evaluating rollout.

Published 24d MCP Developer Tools
✓ No known CVEs patched
Read the diff → Tool health → What is this tool? →

✓ No known CVEs patched in this version

Topics

admin-api ai-gateway llm llmops mcp mcp-server
+3 more
model-context-protocol portkey typescript

Summary

AI summary

Manifest generator now strips PORTKEY_TOOL_DOMAINS/MCP_TOOL_DOMAINS env vars and regenerates LobeHub manifest from live server, fixing truncation and drift.

Full changelog

Fixed

  • migrate_prompt's 0.4.1 description enumerated only two of the three action return values; the common re-run case returns unchanged (prompt found, content identical, no new version created). The description now covers all three outcomes plus the message field.
  • The LobeHub manifest generator inherited the caller's environment, so a stray PORTKEY_TOOL_DOMAINS/MCP_TOOL_DOMAINS export made it silently write a truncated manifest (reproduced: 34 of 150 tools). It now strips both variables, refuses to write fewer tools than the manifest already has, and carries each tool's annotations through to the manifest (previously dropped).
  • Pin the actionlint bootstrap script in CI to a commit SHA — the tag-based raw URL was the one remaining mutable reference in the hardened workflows.
  • Correct two claims in the 0.4.1 changelog entry (mirrored to its GitHub Release): the six tools carry the first hand-authored annotations, not the first annotations, and the manifest drift was 29 tool entries, not 34.

Changed

  • Rewrite the six lowest-scoring tool descriptions (Glama TDQS): update_config, update_integration, create_integration, update_mcp_integration_capabilities, update_mcp_integration_workspaces, and migrate_prompt now disclose side effects and reversibility, name sibling tools and where to source parameter values, and enumerate return fields. These six are also the first tools to carry explicit hand-authored MCP annotations (readOnlyHint/destructiveHint/idempotentHint/openWorldHint), overriding the operation-type-inferred annotations every tool has carried since 0.2.0.

Fixed

  • The LobeHub manifest (lhm.plugin.json) had drifted from the server: 29 tool entries were out of date — the six rewritten descriptions plus 0.4.0's pagination schema changes, including a list_guardrails schema still advertising a page_size maximum of 1000 that the server rejects at 100. The tools array is regenerated from the live server over MCP tools/list (new scripts/generate-lobehub-tools.mjs, wired into publish:lobehub so it can't drift again), which also picks up the Enterprise-gating notes the server appends at registration.
  • glama.json maintainers listed the GitHub org instead of a username, which Glama's claim flow can't match; restored scttbnsn so the org-hosted server page can be claimed.

Full Changelog: https://github.com/CodesWhat/portkey-admin-mcp/compare/v0.4.1...v0.4.2

Weekly OSS security release digest.

The CVE patches and breaking changes that affected production tools this week. One email, every Sunday.

No spam, unsubscribe anytime.

Share this release

Track s-b-e-n-s-o-n/portkey-admin-mcp

Get notified when new releases ship.

Sign up free

About s-b-e-n-s-o-n/portkey-admin-mcp

MCP server for the Portkey AI Gateway Admin API — 150 tools for prompts, configs, analytics, keys, guardrails, integrations, and more.

All releases →

Beta — feedback welcome: [email protected]