Audit MCP server options across integrations; flag unofficial third-party servers
メンテナーはふだん 1 日以内に返信
まだ誰も着手していません。
評価
- 難易度
- 5/5
- 見積もり時間
- 1週間以上
- 初心者へのやさしさ
- 25/100
- issue の種類
- ドキュメント
- 明瞭さ
- おおむね明確
- 活発さ
- 活発
- 技術スタック
- github, javascript, python
調査の方向性
Start with integrations/catalog/*.json and search the repository for other concrete MCP server references; the issue identifies this catalog as the source surfaced by the integrations packages and Customize pane. Cross-check each server’s provenance and health against its repository or package metadata, and check authentication against vendor documentation. Done means a committed audit document inventories every catalog entry, flags risky servers with recommended actions, and links a follow-up issue for each required decision or change.
索引モデルが issue の本文から書いたものです。
説明
Summary
We need a full audit of every MCP server option surfaced in the OpenHands extensions/integrations catalog (and anywhere else we point users at MCP servers). The goal is to identify which integrations currently rely on unofficial third-party MCP servers rather than first-party/official servers provided by the underlying vendor, and flag them for review.
Why
- Partner pushback is already happening. While reviewing our Slack MCP integration, Alianza flagged that the Customize pane linked to
zencoderai/slack-mcp-server— a third-party server onv0.0.1, stale since July 2025, with a non-standard license. They worked around it with a custom server and said they would be "very loathe to interact with that MCP server in production." Slack's official hosted MCP server now exists and resolves that case (see #900). - Unofficial servers carry real risk: unmaintained code, unclear licensing, no vendor accountability, and auth patterns that bypass vendor governance (Slack's official MCP requires workspace-admin approval, audit logs, and Marketplace/internal app identity — third-party servers generally bypass all of that).
- Trust compounds. Enterprise customers screening our integrations (e.g., before adopting OHE) will scrutinize the provenance of every MCP server we reference. One abandoned third-party server undermines confidence in the whole catalog.
- We don't currently know the full blast radius. The Slack case was found by a partner, not by us. There may be more integrations pointing at unofficial servers that we haven't reviewed.
What the audit should do (scope)
For each MCP server option in the catalog:
- Record the underlying server: repo/URL, maintainer, whether it is official (first-party, vendor-published) or third-party/unofficial
- Capture health signals: latest release, last commit/push date, license, stars/activity
- Note the auth model and whether it aligns with the vendor's supported governance model (OAuth, admin approval, etc.)
- Flag any integration whose MCP server is unofficial, abandoned, unclear-licensed, or otherwise risky for production/enterprise use
- Produce a summary table with a recommended action per flag (switch to official server, update pin, remove, keep-with-caveat)
Explicitly out of scope (for this issue)
This issue is only to document the what and why and create the tracking item. The actual investigation of individual integrations and any remediation will happen as follow-up work once the audit plan is agreed.
Related
- #900 — Update Slack MCP integration to Slack's official hosted MCP server (the concrete Slack case that motivated this audit)
Acceptance criteria
- Audit scope and method documented (which surfaces are in scope: integrations catalog, Customize pane, docs, skill references)
- Every MCP server option inventoried with official vs. third-party provenance
- All unofficial/risky servers flagged in a summary table
- Follow-up issues created per flagged integration (remediation/decision needed)
This issue was created by an AI agent (OpenHands) on behalf of the user.
OpenHands AI triage
The following comments and acceptance criteria were added by the OpenHands AI agent.
Triage
This is an audit-and-documentation issue, not a code change, and its scope is resolved from repository evidence. The integrations/catalog/*.json files are the hand-authored source of truth (79 entries) consumed by the @openhands/extensions JS and Python packages and rendered as the Customize pane; today only slack.json references a third-party server package (@zencoderai/slack-mcp-server), which #900 already tracks for remediation. Skills and automation manifests name integrations by id and contain no MCP server references, so no additional surfaces need auditing beyond the catalog itself.
The deliverable is a provenance and health audit document (committed under the existing analysis_results.md pattern) covering all catalog entries: official vs third-party provenance for each underlying MCP server, maintained-server health signals, auth-model fit, and a summary table flagging unofficial, abandoned, unclear-licensed, or otherwise production-risky servers with a recommended action. Remediation changes and the Slack fix are explicitly non-goals and belong to follow-up issues.
The customer-support label and the recorded Alianza partner pushback in this issue and #900 constitute real customer evidence, so this is prioritized medium rather than low.
Acceptance Criteria
- The audit document states the surfaces in scope and the method used, and confirms (with evidence from a repo search) that the hand-authored
integrations/catalog/*.jsonfiles are the only surface referencing concrete MCP servers, plus any adjacent official docs or MCP registries used for cross-checking. - The audit document inventories every entry in
integrations/catalog/*.json(79 files at audit time), each with its underlying MCP server (repo/URL or npm/PyPI package), maintainer, and an official (first-party, vendor-published) vs third-party/unofficial classification that a reviewer can verify against the cited repo or vendor docs. - The audit document records, for each server, at least: latest release/tag, last commit/push date, license identifier, and maintenance/activity signal, sourced from the server's repo or package metadata (not guessed).
- The audit document notes, for each server, the auth model the integration exposes (OAuth, API key/token, unauthenticated) and whether it aligns with the vendor's supported governance model (e.g., vendor OAuth, admin approval, marketplace-app identity), citing the vendor documentation where governance requirements exist.
- The audit document contains a summary table flagging every server that is unofficial, abandoned/stale, unclear-licensed, or otherwise risky for production/enterprise use, with a recommended action per flag (switch to official server, update pin, remove, keep-with-caveat); the Slack entry (
@zencoderai/slack-mcp-server) appears flagged with a cross-reference to #900. - One follow-up GitHub issue exists per integration flagged for a change (remediation or decision needed), each linked from the audit document and to this issue.
- The audit document is committed to the repository and, because it is Markdown-only, requires no catalog or generated-index changes (
npm run build:integrationsandnpm run build:skillsleave no drift).
- 主要言語
- Python
- スター
- 184
- フォーク
- 137
- 平均マージ
- 1日 19時間
- マージ済み PR(30日)
- 46
環境構築
- Dockerfile・Docker Compose ファイルなし
- プルリクエストのテンプレートあり
- コントリビューションガイドなし
はじめの一歩
- issue を最後まで読み、次にプロジェクトのコントリビューションガイドを読みます。
- 着手することを issue にコメントします — 二人が同じ作業をするのを防げます。
- リポジトリをフォークし、ブランチを切って変更します。
- issue 番号を参照したプルリクエストを送ります。
OpenHands/extensions のほかの issue
-
priority:low ready-for-dev
難易度 2/5 1〜3時間 初心者へのやさしさ 68/100
OpenHands/extensions#975 ·
メンテナーはふだん 1 日以内に返信
-
priority:low ready-for-dev
難易度 2/5 1〜3時間 初心者へのやさしさ 74/100
OpenHands/extensions#973 ·
メンテナーはふだん 1 日以内に返信
-
難易度 2/5 1〜3時間 初心者へのやさしさ 76/100
OpenHands/extensions#970 ·
メンテナーはふだん 1 日以内に返信
-
難易度 2/5 1〜3時間 初心者へのやさしさ 78/100
OpenHands/extensions#969 ·
メンテナーはふだん 1 日以内に返信
-
Add a `maven` skill: agents install Debian's `maven` package, which can't load plugins built for Java 17+対応中かも @sumitvairagar が 3 日前に担当しました。 オープンpriority:medium ready-for-dev
難易度 2/5 1〜3時間 初心者へのやさしさ 80/100
OpenHands/extensions#963 ·
メンテナーはふだん 1 日以内に返信
OpenHands/extensions の issue をすべて見る
似ている issue
-
難易度 2/5 1〜3時間 初心者へのやさしさ 78/100
huggingface/datasets#9128 ·
メンテナーはふだん 1 日以内に返信
-
Function/dict state handlers can't supply provider-state injected values (`generate.provider_state`)オープン
難易度 2/5 1〜3時間 初心者へのやさしさ 78/100
pact-foundation/pact-python#1789 ·
メンテナーはふだん 9 日以内に返信
-
documentation
難易度 2/5 1〜3時間 初心者へのやさしさ 82/100
juandresrodca/AdminForge#66 ·
メンテナーはふだん 1 日以内に返信
-
難易度 2/5 1〜3時間 初心者へのやさしさ 68/100
メンテナーはふだん 1 日以内に返信
-
backend/local comp/tools P2 tool/terminal type/bug
難易度 2/5 1〜3時間 初心者へのやさしさ 72/100
NousResearch/hermes-agent#158788 ·
メンテナーはふだん 1 日以内に返信