Skip to main content
Glama

Lab Registry Server

MCP server exposing the Gen-e2 Lab Registry — skills, agents, commands, and hooks — to any MCP-compatible client (Claude Code, GitHub Copilot agent mode). Designed for the Innovation Lab at Palo IT Singapore.

Canonical repository: https://github.com/Palo-IT-GitHub-Demos/lab-registry-mcp

This repository contains only the MCP server. It does not embed the gen-e2-marketplace project, which remains the registry source of truth. In normal usage, the server reads that source directly from GitHub.


What it does

This MCP server connects to the gen-e2 plugins repository and gives agents direct access to the library from VS Code.

Who it's for

It is mainly intended for Labs developers, especially those who are new to gen-e2.

Why it exists

It makes the gen-e2 library easier to discover, reuse, and integrate across projects.

Main use cases

  • Explore the gen-e2 artefact library in more depth

  • Find a specific plugin or artefact by name and integrate it easily into the current project

  • Get suggestions for which gen-e2 artefacts are most relevant to the current project

  • Stay informed about new gen-e2 plugins and updates to existing ones


Related MCP server: Taproom

How it works

gen-e2-marketplace/          ← source of truth (read-only, never written)
  .claude-plugin/
    marketplace.json         ← list of all 13 plugins with semver versions
  plugins/
    android/
      .claude-plugin/plugin.json
      CHANGELOG.md
      skills/android-architecture/SKILL.md   ← YAML frontmatter + markdown body
      agents/android-architect.agent.md
      commands/add-screen.md
      hooks.json
    research-suite/ ...
    delivery/ ...
    ...

lab-registry-mcp/            ← this repo
  src/lab_registry/
    registry.py              ← reads marketplace on first call, caches result
    models.py                ← RegistryEntry, Plugin (Pydantic)
    server.py                ← FastMCP, 15 tools registered via @mcp.tool()
    tools/
      search.py              ← list_entries, search_entries, suggest_entries, suggest_plugins
      fetch.py               ← get_entry, get_plugin, get_entry_by_id, list_plugins, get_changelog
      compliance.py          ← check_compliance, check_compliance_plugin
      stats.py               ← get_marketplace_stats
      validate.py            ← validate_entry

Startup sequence:

  1. MCP client (Claude Code or Copilot) spawns the server process via stdio

  2. Server responds to initialize — no files read yet

  3. On first tool call, load_registry() reads from either REGISTRY_GITHUB_REPO or REGISTRY_PATH, parses marketplace metadata, indexes plugin entries, and extracts updated_at from CHANGELOG.md when available

  4. Result is cached in memory (lru_cache) for the life of the process

  5. All subsequent tool calls use the in-memory index — no disk access except get_entry (reads file content on demand)

Versioning model: version lives at the plugin level (from plugin.json), not per individual artifact. All 33 android artefacts share plugin_version: "0.1.0". If the android plugin bumps to 0.2.0, all its artefacts are considered outdated.


Install

# Install directly from GitHub
pip install "git+https://github.com/Palo-IT-GitHub-Demos/lab-registry-mcp"

# Or pin the current release tag
pip install "git+https://github.com/Palo-IT-GitHub-Demos/lab-registry-mcp@v0.3.1"

For local development:

git clone https://github.com/Palo-IT-GitHub-Demos/lab-registry-mcp
cd lab-registry-mcp

python3 -m venv .venv
source .venv/bin/activate
pip install -e ".[dev]"

Configuration

There are 3 practical ways to configure the source registry.

1) GitHub source from GLOBAL-PALO-IT/gen-e2-marketplace (recommended when you have access)

Use this when you can obtain a GitHub token with access to the official marketplace repository.

export REGISTRY_GITHUB_REPO=GLOBAL-PALO-IT/gen-e2-marketplace
# optional but usually needed for private repo access
export REGISTRY_GITHUB_TOKEN=ghp_...

2) GitHub source from your own fork (temporary workaround)

Use this when access to the official repo token is difficult, but you can fork the marketplace into a personal or easier-to-access repository.

export REGISTRY_GITHUB_REPO=<your-user-or-org>/gen-e2-marketplace
export REGISTRY_GITHUB_TOKEN=ghp_...

3) Local source from a clone of gen-e2-marketplace

Use this for offline development, local debugging, or local integration/E2E tests.

export REGISTRY_PATH=/abs/path/to/gen-e2-marketplace
# or copy .env.example → .env and set it there

Client setup

Both Claude Code and Copilot use the same stdio server; only the client registration format changes.

Claude Code CLI — user-level

claude mcp add lab-registry --scope user \
  -e REGISTRY_GITHUB_REPO=GLOBAL-PALO-IT/gen-e2-marketplace \
  -e REGISTRY_GITHUB_TOKEN=ghp_... \
  -- /abs/path/to/.venv/bin/mcp run /abs/path/to/src/lab_registry/server.py

If you use a fork or a local clone, replace the env vars accordingly.

GitHub Copilot agent mode — ~/Library/Application Support/Code/User/mcp.json

{
  "servers": {
    "lab-registry": {
      "type": "stdio",
      "command": "/abs/path/to/.venv/bin/mcp",
      "args": ["run", "/abs/path/to/src/lab_registry/server.py"],
      "env": {
        "REGISTRY_GITHUB_REPO": "GLOBAL-PALO-IT/gen-e2-marketplace",
        "REGISTRY_GITHUB_TOKEN": "ghp_..."
      }
    }
  }
}

For local mode, replace the env block with:

{
  "REGISTRY_PATH": "/abs/path/to/gen-e2-marketplace"
}

Run

# Visual debug UI (MCP Inspector at http://localhost:6274)
mcp dev src/lab_registry/server.py

# Stdio (for client config)
REGISTRY_GITHUB_REPO=owner/repo REGISTRY_GITHUB_TOKEN=ghp_... mcp run src/lab_registry/server.py

# Unit + integration + E2E tests
REGISTRY_PATH=../gen-e2-marketplace pytest tests/ -q

The 15 tools

Response shapes: list_entries, search_entries, and suggest_entries use a lean serialization — only populated fields are returned (no null or empty-list noise). get_entry always returns the full entry shape.

Tool selection guide

Goal

Use

Discover which plugins fit a project type

suggest_plugins

Exact keyword or partial name (e.g. "tdd", "android")

search_entries

Natural language task description (e.g. "I need to review architecture")

suggest_entries

Browse by type or plugin

list_entries with filters

Install a full plugin (all artefacts + paths)

get_plugin_install_package

Read one specific artefact

get_entry or get_entry_by_id

Check installed plugins are up to date (one plugin)

check_compliance_plugin

Check installed plugins are up to date (multiple)

discover plugin.json → check_compliance

See what changed in a plugin

get_changelog

list_entries

List all registry entries. Returns a flat list of RegistryEntry objects.

Parameter

Type

Description

type

string?

Filter: "skill", "agent", "command", or "hook"

plugin

string?

Filter by plugin name (e.g. "android")

tags

string[]?

Filter by keywords — OR match (any tag must match)

// Example: all skills in the android plugin
{ "type": "skill", "plugin": "android" }

search_entries

Keyword search over name, description, and plugin name. Name matches are ranked above description matches.

Parameter

Type

Description

query

string

Search term

type

string?

Optional type filter

{ "query": "architecture", "type": "skill" }

suggest_entries

Task-oriented suggestion tool. Splits the task into individual terms and scores entries by how many terms appear in their name, description, plugin name, and tags.

Use for natural language queries at the artefact level. For plugin-level discovery ("which plugins fit my project?"), use suggest_plugins instead. For exact keyword matching, use search_entries.

Parameter

Type

Description

task

string

Natural language description of what you need

type

string?

Optional type filter

limit

integer?

Maximum number of results (default 5)

{ "task": "I need to write tests for a Go service", "type": "skill", "limit": 5 }

suggest_plugins

Plugin-level discovery. Scores plugins by how many task words appear in their name (3× weight), description, and tags.

Use this before suggest_entries when the user describes their project type and wants to know which plugins are most relevant — rather than listing all 15 plugins flat.

Parameter

Type

Description

task

string

Natural language description of the project or need

limit

integer?

Maximum number of results (default 5)

{ "task": "I'm building an Android app", "limit": 3 }

Example response: android, delivery, architecture-reviewer — each with score and matched_terms.


get_entry

Fetch the full content of a specific entry. Returns structured metadata and the raw markdown body.

Parameter

Type

Description

plugin

string

Plugin name

type

string

Artifact type

name

string

Artifact name

{ "plugin": "android", "type": "skill", "name": "android-architecture" }

Response shape:

{
  "entry":          { "id": "android/skill/android-architecture", "plugin_version": "0.1.0", ... },
  "metadata":       { "name": "android-architecture", "description": "..." },
  "content_raw":    "# Android architecture (project delta)\n\n...",
  "content_full":   "---\nname: android-architecture\n---\n# Android architecture...\n",
  "install_targets": {
    "claude_local":    ".claude/skills/android-architecture/SKILL.md",
    "copilot":         ".github/skills/android-architecture/SKILL.md",
    "plugin_tracking": ".claude/plugins/android/plugin.json"
  }
}

content_full is the verbatim source file (frontmatter + body) — write it directly to install_targets.claude_local or install_targets.copilot without any reconstruction.


get_entry_by_id

Fetch one entry directly from its canonical ID. Returns the same shape as get_entry: entry, metadata, content_raw, content_full, and install_targets.

Parameter

Type

Description

id

string

Entry ID in plugin/type/name format

{ "id": "android/skill/android-architecture" }

get_plugin

All entries for one plugin, plus its manifest.

Parameter

Type

Description

plugin

string

Plugin name

{ "plugin": "research-suite" }

Response: { "manifest": { "version": "1.0.1", ... }, "entries": [...] }


list_plugins

List all indexed plugins with version and per-type entry counts.

Response includes plugin-level summary fields such as version, updated_at, and counts for skills, agents, commands, and hooks.


get_changelog

Return the raw CHANGELOG.md content for a plugin.

Parameter

Type

Description

plugin

string

Plugin name

{ "plugin": "delivery" }

get_marketplace_stats

Return marketplace-level statistics: totals, counts by type, counts by plugin, and latest update information.

Includes last_updated (most recent date across all entries) and last_updated_plugin (name of the plugin that was updated most recently).

Useful for dashboards, summaries, and quick health checks.


check_compliance

Check whether locally installed gen-e2 plugin artefacts are up to date with the registry.

For a single plugin, use check_compliance_plugin instead — it requires only the plugin name and local version, without listing artefacts manually.

Recommended workflow for multiple plugins: discover local plugin.json files, read the version field from each, build the entries list, then call this tool.

Each item in entries must have name, type, plugin, and local_version.

{
  "entries": [
    { "name": "research",   "type": "skill", "plugin": "research-suite", "local_version": "0.8.0" },
    { "name": "coi-verify", "type": "skill", "plugin": "research-suite", "local_version": "0.8.0" }
  ]
}

Response:

{
  "outdated": [{ "name": "research", "plugin": "research-suite", "local_version": "0.8.0", "registry_version": "1.0.1" }],
  "unknown":  [],
  "up_to_date_count": 0
}

outdated = version mismatch. unknown = not found in registry.


check_compliance_plugin

Shortcut: check all artefacts of a plugin against a single local version in one call.

Equivalent to calling get_plugin to list artefacts, then check_compliance with each one. Use this when you have a plugin.json with one version field — it removes the need to enumerate artefacts manually.

Parameter

Type

Description

plugin

string

Plugin name

local_version

string

Version from the local plugin.json

{ "plugin": "research-suite", "local_version": "0.8.0" }

Returns the same shape as check_compliance: outdated, unknown, up_to_date_count.

Validate a skill, agent, or command markdown file structure against the expected schema.

Typical output includes:

  • valid

  • errors

  • warnings

  • parsed frontmatter when available

Useful before contributing a new artefact to the marketplace.


get_plugin_install_package

Return a complete install package for a plugin — one call, everything needed to install.

Parameter

Type

Description

plugin

string

Plugin name

{ "plugin": "implementation-plan" }

Response shape:

{
  "plugin": { "name": "implementation-plan", "version": "0.1.0", ... },
  "files": [
    {
      "id": "implementation-plan/skill/create-implementation-plan",
      "type": "skill",
      "name": "create-implementation-plan",
      "content_full": "---\nname: create-implementation-plan\n---\n...",
      "install_targets": {
        "claude_local":    ".claude/skills/create-implementation-plan/SKILL.md",
        "copilot":         ".github/skills/create-implementation-plan/SKILL.md",
        "plugin_tracking": ".claude/plugins/implementation-plan/plugin.json"
      }
    }
  ],
  "plugin_tracking": {
    "path":    ".claude/plugins/implementation-plan/plugin.json",
    "content": "{\"name\": \"implementation-plan\", \"version\": \"0.1.0\", ...}"
  }
}

Prefer this over multiple get_entry calls when installing a full plugin. Each file in files has content_full (write-ready) and install_targets (exact paths per client).


reload_registry

Force reload the in-memory cache from its source (GitHub or local). Use after a marketplace update to get fresh data without restarting the server.

Response: { "added": [...], "removed": [...], "modified": [...], "total": N }


Usage examples

These are natural-language prompts validated against the live registry.

1) Discover what exists

What gen-e2 plugins are available and which was updated most recently?

Typical tools used: get_marketplace_stats (returns last_updated_plugin) + list_plugins


2) Find by type

What gen-e2 agents are available in the registry?

Typical tools used: list_entries with type="agent"


Search the gen-e2 registry for skills related to architecture review.

Typical tools used: suggest_entries with a task description


4) Get documentation and install files

Get the full documentation and install files for the gen-e2 delivery plugin.

Typical tools used: get_plugin_install_package — returns all artefacts with content_full + install_targets in one call


5) Install a plugin into the current project

Install the gen-e2 implementation-plan plugin into my project for both Claude Code and GitHub Copilot.

Typical tools used: get_plugin_install_package → write each file.content_full to file.install_targets.claude_local and file.install_targets.copilot


6) Check for outdated plugins

Check if my locally installed gen-e2 plugins are up to date with the registry.

Typical tools used: discover .claude/plugins/*/plugin.json → check_compliance → get_changelog for outdated entries


7) Read one entry

Get the full content of android/skill/android-architecture

Typical tools used: get_entry_by_id


8) Validate a new contribution

Validate this new skill markdown file against the gen-e2 schema

Typical tools used: validate_entry


9) Refresh cache after marketplace updates

Reload the gen-e2 registry and tell me what changed

Typical tools used: reload_registry


Registry coverage

Current state of gen-e2-marketplace as indexed:

Plugin

Version

Skills

Agents

Commands

Hooks

android

0.1.0

14

9

9

1

architecture-reviewer

0.1.0

4

1

0

0

delivery

0.2.3

5

1

0

0

figma-design-to-code

0.1.0

1

0

0

0

fortran77-explainer

0.1.0

0

1

0

0

go-tdd-orchestrator

0.1.0

2

3

0

0

html-planner-and-presentation

0.1.0

3

0

0

0

html-presentation

0.1.0

1

0

0

0

implementation-plan

0.1.0

1

0

0

0

kotlin-and-kotlin-multiplatform

0.1.0

4

0

0

0

migration-implementation-plan

0.1.0

1

0

0

0

research-suite

1.0.1

2

0

0

0

swift5-development-test-writer

0.1.0

2

0

0

0

Total

40

15

9

1

65 artefacts indexed across 13 plugins when reading from GLOBAL-PALO-IT/gen-e2-marketplace.

dev-workflow exists in a local clone of the marketplace but is not published to the shared GitHub repository — it is excluded from the counts above. updated_at is null for plugins without a CHANGELOG.md.


Tests

With RUN_NL_TESTS unset, pytest --collect-only -q collects 193 tests. The NL dispatch module is excluded unless RUN_NL_TESTS=1 and its Ollama health check succeeds.

Test level

Setup

Command

Measured result

Unit, contract, and mocked GitHub

No environment variables

pytest tests/ -q

158 passed, 36 skipped; integration/E2E and NL are skipped

Integration and E2E

REGISTRY_PATH points to a local marketplace

REGISTRY_PATH=../gen-e2-marketplace pytest tests/ -q

193 passed, 1 skipped; NL module is skipped

NL dispatch

RUN_NL_TESTS=1, REGISTRY_PATH, and healthy Ollama

RUN_NL_TESTS=1 REGISTRY_PATH=../gen-e2-marketplace pytest tests/test_nl_dispatch.py -v -m nl_dispatch

Optional; excluded from the default collection

The standard summary's 36 skips are 35 individual integration/E2E cases plus one module-level NL skip; that module adds no test item to the 193 collected tests.

The standard profile uses a temporary fixture and mocked GitHub HTTP; it does not require network access. The local-registry profile adds real marketplace integration and MCP subprocess scenarios. NL dispatch uses Ollama only after the explicit opt-in and a successful 500 ms /api/tags health check.


Known limitations

  • No per-artifact versioning — version is at plugin level; a plugin bump marks all its artefacts as outdated even if only one changed

  • No deprecated detection — no deprecated flag in the source format

  • updated_at is best-effort — parsed from CHANGELOG.md; null if absent

  • Cache on demand — use reload_registry tool to refresh without restarting the server

  • Hooks indexed one entry per plugin — not per event type

Available Tools

15 tools
check_complianceA

Check whether locally installed gen-e2 plugin artefacts are up to date with the registry.

Use this whenever the user asks to verify, check, or audit their installed gen-e2 plugins. Workflow: discover local plugin.json files (e.g. .claude/plugins/*/plugin.json), read the version field from each, then call this tool — do NOT compare versions manually.

Each item in entries must have:

  • name: artifact name (e.g. "research", "commit-push-pr")

  • type: "skill", "agent", "command", or "hook"

  • plugin: plugin name (e.g. "research-suite", "delivery")

  • local_version: version read from the local plugin.json (e.g. "0.8.0")

Returns:

  • outdated: entries where local_version != current registry version (includes registry_version)

  • unknown: entries not found in the registry

  • up_to_date_count: number of entries that match the registry

Inputs: entries (required). Returns: object

check_compliance_pluginA

Check all artefacts of a gen-e2 plugin against a single local version.

Use this when you have a plugin.json with one version field and want to verify the whole plugin in one call, instead of listing artefacts manually and calling check_compliance with each one.

Typical workflow:

  1. Read .claude/plugins//plugin.json → get local version

  2. Call check_compliance_plugin(plugin=, local_version=)

Returns the same shape as check_compliance:

  • outdated: artefacts where local_version != registry version (includes registry_version)

  • unknown: artefacts not found in the registry

  • up_to_date_count: number of matching artefacts

Inputs: plugin (required), local_version (required). Returns: object

get_changelogA

Get the full CHANGELOG.md content for a plugin.

Useful after check_compliance signals an outdated entry — shows what changed between versions without leaving the MCP context.

Returns:

  • plugin: plugin name

  • version: current version

  • changelog_raw: full CHANGELOG.md text (null if absent)

Inputs: plugin (required). Returns: object

get_entryA

Get a specific registry entry with its full content.

Use this ONLY when you already know the exact plugin, type, AND name of the artefact. If you only have a task description or keyword, use suggest_entries or search_entries instead to discover the right artefact first. To get all artefacts of a plugin at once, use get_plugin_install_package instead.

Returns:

  • entry: structured RegistryEntry fields

  • metadata: parsed YAML frontmatter from the source file

  • content_raw: markdown body below the frontmatter (the actual instructions)

  • content_full: verbatim file content (frontmatter + body), write-ready

  • install_targets: exact destination paths per client (claude_local, copilot, plugin_tracking)

Inputs: name (required), type (required), plugin (required). Returns: object

get_entry_by_idA

Get a registry entry by its full ID (plugin/type/name).

Shortcut for get_entry when you already have the entry ID from list_entries. Returns same shape as get_entry: {entry, metadata, content_raw}.

Example: id="android/skill/android-architecture"

Inputs: id (required). Returns: object

get_marketplace_statsB

Get a dashboard overview of the entire registry.

Returns:

  • total_entries, total_plugins

  • by_type: entry counts per artifact type

  • by_plugin: per-plugin breakdown with version and counts

  • last_updated: most recent CHANGELOG date across all plugins

  • plugins_without_changelog: plugins missing a CHANGELOG.md

Inputs: no named parameters. Returns: object

get_pluginC

Get all registry entries for a plugin, plus its manifest.

Returns:

  • manifest: Plugin metadata (name, version, description, tags)

  • entries: list of all RegistryEntry objects in the plugin

Inputs: plugin (required). Returns: object

get_plugin_install_packageA

Get full documentation AND install-ready files for an entire plugin in one call.

Use this whenever the user wants to:

  • install a plugin into their project

  • get the documentation and install files for a plugin

  • know where to place a plugin's artefacts in their project

Returns all artefacts with:

  • content_full: verbatim file content (frontmatter + body), write-ready — use this to write the file directly without any reconstruction

  • install_targets: exact destination paths per client: claude_local → .claude/skills|agents|commands/{name}/... copilot → .github/skills|agents|prompts/{name}/... plugin_tracking → .claude/plugins/{plugin}/plugin.json

  • plugin_tracking: the plugin.json content to write for compliance tracking

Prefer this over multiple get_entry calls when working with a full plugin.

Inputs: plugin (required). Returns: object

list_entriesA

List registry entries, optionally filtered by type, plugin, or tags.

Use type="agent" to list agents, type="skill" to list skills, type="command" to list commands, type="hook" to list hooks. Omit all filters to list every entry in the registry.

Filter by:

  • type: "skill", "agent", "command", or "hook"

  • plugin: plugin name (e.g. "android", "delivery", "research-suite")

  • tags: list of keywords (OR match — any tag must match)

Inputs: tags, type, plugin. Returns: result

list_pluginsA

List all plugins with version, description, and entry counts per type.

Returns one entry per plugin sorted alphabetically, with:

  • name, version, description, tags, author

  • entry_counts: {skill, agent, command, hook, total}

Inputs: no named parameters. Returns: result

reload_registryA

Force reload the registry from its source (GitHub or local path).

Clears the in-memory cache and re-fetches all entries. Use after a marketplace update (git pull / new commit) to get fresh data without restarting the server.

Returns a diff vs the previous state:

  • added: entry IDs new in this reload

  • removed: entry IDs no longer present

  • modified: entry IDs whose plugin_version changed

  • total: total number of entries after reload

Inputs: no named parameters. Returns: object

search_entriesA

Search registry entries by exact keyword or partial name match.

Searches name, description, and plugin name. Name matches are ranked before description matches. Optionally restrict results to a specific type.

Use this when you have a specific keyword or partial name (e.g. "tdd", "android", "architecture"). For natural language task descriptions, use suggest_entries instead.

Inputs: query (required), type. Returns: result

suggest_entriesA

Suggest registry entries relevant to a natural language task description.

Use this when you do NOT have a specific artefact name — only a task or intent description. It scores entries by term overlap and returns ranked suggestions. If you already know the exact plugin+type+name, use get_entry instead. For exact keyword or partial name matching, use search_entries instead.

Examples: "I need to review architecture and create ADRs", "write tests for a Go service", "build an Android feature with MVVM".

  • task: natural language description of what you need

  • type: optional type filter ("skill", "agent", "command", "hook")

  • limit: max results to return (default 5)

Inputs: task (required), type, limit. Returns: result

suggest_pluginsA

Suggest gen-e2 plugins relevant to a natural language task or project description.

Use this for plugin-level discovery when the user describes their project or use case and wants to know which plugins are most relevant — before drilling into individual artefacts with suggest_entries.

Scores plugins by how many task words appear in their name (3x weight), description, and tags. Returns plugins sorted by relevance score.

Examples:

  • "I'm building an Android app" → android, delivery, architecture-reviewer

  • "I need to research and document a technical decision" → research-suite, delivery

  • "Go microservice with TDD" → go-tdd-orchestrator

  • task: natural language description of the project or need

  • limit: max results to return (default 5)

Inputs: task (required), limit. Returns: result

validate_entryA

Validate a skill/agent/command markdown file against the marketplace schema.

Checks required fields, recommended fields, body content, and naming conventions.

  • content: full file content including YAML frontmatter

  • type: "skill", "agent", "command", or "hook"

Returns:

  • valid: bool

  • errors: blocking schema violations

  • warnings: non-blocking recommendations

  • parsed: the parsed YAML frontmatter

Inputs: type (required), content (required). Returns: object

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A3.9/5.0

    Scored across 15 tools

    Disambiguation4/5

    Most tools target distinct operations (list/search/suggest entries vs plugins, compliance, validation, reload), and descriptions explicitly cross-reference when to use alternatives. However, several near-duplicate pairs exist (get_entry vs get_entry_by_id, get_plugin vs get_plugin_install_package, check_compliance vs check_compliance_plugin) that could be misselected despite the guidance.

    Naming Consistency5/5

    Every tool follows a consistent snake_case verb_noun or noun pattern (get_* , list_*, check_*, search_*, suggest_*, reload_*, validate_*). No camelCase mixing or vague verbs; the longest names still conform to the same convention.

    Tool Count4/5

    15 tools sit at the top of the well-scoped range and each maps to a plausible registry operation. A few are convenience wrappers (get_entry_by_id, check_compliance_plugin) that slightly inflate the count, but nothing feels gratuitous.

    Completeness5/5

    The surface covers the full lifecycle of a read-only marketplace: discovery (list/search/suggest for both entries and plugins), retrieval (entries, plugins, install packages), compliance checking, changelog, stats, schema validation, and cache reload. No obvious operational gaps remain for the agent's workflows.

    Maintenance

    ActivitySlowing
    ResponsivenessNo recent community issues

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      Not graded
      quality
      C
      maintenance
      Enables AI agents to discover, install, and manage SKILL.md skills from a Git-backed registry via MCP tools for search, install, and list operations.
      35 npm
      2
      MIT
    • A
      license
      Not graded
      quality
      D
      maintenance
      Discovers and manages portable agent capabilities (skills and MCP servers) from configurable collections, providing search, inspection, and local installation via CLI and MCP tools.
      1
      MIT
    • A
      license
      Not graded
      quality
      C
      maintenance
      Enables managing a canonical library of agent skills and MCP servers, syncing them across multiple development harnesses, and adding, importing, or configuring them through MCP tools.
      92 npm
      MIT
    • A
      license
      A
      quality
      B
      maintenance
      Enables MCP hosts to discover, browse, and load Agent Skills from one or many SKILL.md libraries over the Model Context Protocol, with namespaced libraries, search/list/get tools, progressive disclosure, and a digest-verified pull client for syncing skills to disk.
      7
      1
      MIT