Skills & Extensions

Context7 Docs Lookup

Up-to-date library docs and code examples: resolve a library id, then query its docs.

Why use it:Built-in skills are offline teaching content; Context7 covers the latest docs and API changes published after model training cutoffs.

Install

1. Settings → Extensions → "Add Extension", type STDIO
2. Command: npx -y @upstash/context7-mcp
3. No API key required
4. Enable the toggle and start a new session

Before you install

Credentials:No credentials required (free).

Data flow:Library names and doc queries are sent to the Context7 service; results return to the session.

Exa AI Search

AI-powered search with neural and keyword modes for web search and content fetch.

Why use it:Interchangeable with Tavily; Exa specializes in semantic (embedding-based) search rather than keyword matching.

Install

1. Settings → Extensions → "Add Extension", type STDIO
2. Command: npx -y exa-mcp-server
3. Add environment variable EXA_API_KEY (from exa.ai)
4. Enable the toggle and start a new session

Before you install

Credentials:Requires EXA_API_KEY (from exa.ai); stored only in your local system keychain.

Data flow:Queries are sent to the Exa cloud API; results return to the session.

Excalidraw Diagrams

Create and edit Excalidraw diagrams in-session: flowcharts, architecture sketches and whiteboards.

Why use it:Built-in visualisation covers chart generation; Excalidraw adds an editable canvas with hand-drawn style diagrams.

Install

{"mcpServers":{"excalidraw":{"url":"https://excalidraw-mcp-app.vercel.app/mcp"}}}

Paste into Settings → Extensions → "Add Extension" (type: Streamable HTTP), or edit config.yaml manually.

Dynamic source: hosted endpoint or unpinned package — behavior may change with upstream updates.

Before you install

Credentials:No credentials required.

Data flow:Diagram descriptions are sent to excalidraw-mcp-app.vercel.app and results return via that service; access from mainland China requires a proxy.

GitHub (Official)

Official GitHub remote extension: repository management, issues, pull requests, code search and analysis.

Why use it:Lets the agent work your GitHub workflow directly (issues, PR reviews, code search); local git operations stay with the built-in developer tools.

Install

{"mcpServers":{"github":{"url":"https://api.githubcopilot.com/mcp","headers":{"Authorization":"Bearer <YOUR_PAT>"}}}}

Paste into Settings → Extensions → "Add Extension" (type: Streamable HTTP), or edit config.yaml manually.

Dynamic source: hosted endpoint or unpinned package — behavior may change with upstream updates.

Before you install

Credentials:Requires a GitHub Personal Access Token (least-privilege scope recommended); passed via Authorization header, stored locally.

Data flow:Requests go through api.githubcopilot.com (GitHub-hosted endpoint).

MongoDB Database

Official MongoDB extension: queries, aggregations, collection and index management — connected to your own Atlas cloud or local instance.

Why use it:Built-in developer tools handle local files and commands; the MongoDB extension connects the agent to your database — queries, aggregations and index diagnosis right in the session.

Install

1. Settings → Extensions → "Add Extension", type STDIO
2. Command: npx -y mongodb-mcp-server (read-only first recommended: append --readOnly)
3. Add environment variable MDB_MCP_CONNECTION_STRING with your connection string (from the Atlas console; local instance e.g. mongodb://localhost:27017)
4. Enable the toggle and start a new session

Before you install

Credentials:Requires a MongoDB connection string (from the Atlas console, or a local instance mongodb://localhost:27017); passed via environment variable, stored locally.

Data flow:Queries and results round-trip between you and the connected MongoDB instance (Atlas or local); no other third-party service involved.

Scholar Sidekick Citations

Academic citation toolkit: resolve DOI / PMID / ISBN / arXiv ids, format citations, check retractions and open-access status.

Why use it:Built-in tools do not verify academic citations; this adds retraction warnings and open-access checks for writing and literature review.

Install

1. Settings → Extensions → "Add Extension", type STDIO
2. Command: npx -y scholar-sidekick-mcp@latest
3. No API key required
4. Enable the toggle and start a new session

Dynamic source: hosted endpoint or unpinned package — behavior may change with upstream updates.

Before you install

Credentials:No credentials required (free).

Data flow:Identifiers and citation data go through scholar-sidekick.com and public scholarly APIs.

Tavily Web Search

Web search and content extraction: search, extract, crawl and site mapping in one tool call.

Why use it:The built-in browser interacts with pages you point it at; Tavily adds web-wide search and structured extraction without controlling your browser.

Install

1. Open Settings → Extensions → "Add Extension", type STDIO
2. Command: npx -y tavily-mcp
3. Add environment variable TAVILY_API_KEY with your key (free at tavily.com)
4. Enable the toggle and start a new session

Before you install

Credentials:Requires TAVILY_API_KEY (free at tavily.com); stored only in your local system keychain.

Data flow:Queries are sent to the Tavily cloud API; results return to the session.

YouTube Transcripts

Fetch YouTube transcripts and video info, turning video content into analyzable text.

Why use it:The built-in browser renders pages but cannot read caption tracks; this turns video content into text for summarizing, translating and quoting.

Install

1. Prerequisite: install uv (official installer, or pip install uv)
2. Settings → Extensions → "Add Extension", type STDIO
3. Command: uvx --from git+https://github.com/jkawamoto/mcp-youtube-transcript mcp-youtube-transcript
4. No API key; enable the toggle and start a new session (first build is slower, then cached)

Dynamic source: hosted endpoint or unpinned package — behavior may change with upstream updates.

Before you install

Credentials:No credentials required (free).

Data flow:Video ids are sent to YouTube; transcript text returns to the session. Package code is pulled from GitHub.