MCP Servers

Hand-maintained catalog of MCP servers configured in this environment, organized by category.

49 configured servers

What is MCP?

The Model Context Protocol is an open standard for connecting AI agents to external tools and data sources. Each server below exposes one or more tools that agents can invoke during a session.

Source of truth for this page

This page is hand-maintained and currently summarizes the MCP server entries in the repository root mcp.json. That file includes a mix of local scripts, local packages, remote gateways, and third-party servers, so this list is broader than the repo-local mcp/ directory.

Managed client sync coverage

Sync source of truth

Edit config/mcp-registry.json, then preview local drift with uv run python scripts/sync_agent_stack.py --targets home --check. Apply repo and home updates with uv run python scripts/sync_agent_stack.py --targets repo,home --apply.

MCPHub local control plane

MCPHub is now the preferred local MCP control plane. When mcphub.enabled is true in config/mcp-registry.json, clients receive MCPHub all, group, server, and disabled Smart Routing endpoints rooted at http://127.0.0.1:46683. Start it locally with just mcphub-up; Docker is not used for the managed launch path. ChatGPT remote MCP access uses the opt-in Cloudflare Tunnel URL from .env.mcphub.

Exact vs merge behavior

ChatGPT Desktop, the GitHub Copilot shadow config, Gemini antigravity, the Gemini extension config, OpenCode, AITK, and Crush all receive the canonical baseline and preserve non-canonical extras already present in those files. Claude Desktop, Cursor, Gemini CLI, GitHub Copilot CLI, and Codex are rewritten to the exact canonical baseline. Cherry Studio receives a generated import pack in a tool-owned subdirectory instead of direct config mutation so unmanaged local MCP entries and ad hoc import bundles in the parent directory stay untouched.

Client	Managed path	Root key / shape	Sync behavior
ChatGPT Desktop	~/Library/Application Support/ChatGPT/mcp.json	mcpServers	Canonical baseline + preserve non-canonical extras
Claude Code local enablement	~/.claude/settings.local.json	enabledMcpjsonServers	Filters enabled names to valid synced servers
Claude Desktop	~/Library/Application Support/Claude/claude_desktop_config.json	mcpServers	Exact canonical baseline
Cherry Studio	~/Library/Application Support/CherryStudio/mcp-import/managed/	all.json + one <server>.json per managed server	Generated import pack; does not mutate live app state
Codex	~/.codex/config.toml	[mcp_servers.<name>]	Exact canonical baseline
Crush	~/.config/crush/crush.json	mcp	Canonical baseline + preserve non-canonical extras
Cursor	~/.cursor/mcp.json	mcpServers	Exact canonical baseline
Gemini CLI	~/.gemini/settings.json	mcpServers	Exact canonical baseline
Gemini antigravity	~/.gemini/antigravity/mcp_config.json	mcpServers	Canonical baseline + preserve non-canonical extras
Gemini extension antigravity	~/.gemini/extensions/outline-driven-development/antigravity/mcp_config.json	mcpServers	Canonical baseline + preserve extension-specific extras
GitHub Copilot CLI	~/.copilot/mcp-config.json	mcpServers	Exact canonical baseline with Copilot tool allowlists
GitHub Copilot shadow config	~/.config/.copilot/mcp-config.json	mcpServers	Canonical baseline + preserve non-canonical extras in Copilot-style schema
OpenCode	~/.config/opencode/opencode.json	mcp	Canonical baseline + preserve non-canonical extras in local/remote schema
AITK	~/.aitk/mcp.json	servers	Canonical baseline + preserve non-canonical extras

1. Update the canonical MCP registry.

   Edit config/mcp-registry.json or the repo sources that feed it.

2. Preview repo and home drift.

   uv run python scripts/sync_agent_stack.py --targets repo,home --check

3. Apply the synchronized config set.

   uv run python scripts/sync_agent_stack.py --targets repo,home --apply

4. Import Cherry Studio MCP files only if you use Cherry Studio.

   Use ~/Library/Application Support/CherryStudio/mcp-import/managed/<server>.json for additive one-server imports, or ~/Library/Application Support/CherryStudio/mcp-import/managed/all.json only when you intentionally want to replace Cherry Studio’s full managed MCP list.

   Top-level files already sitting in ~/Library/Application Support/CherryStudio/mcp-import/ are left alone intentionally.

DuckDuckGo setup (representative client snippets)

Use server key duckduckgo with the duckduckgo-mcp-server package.

Standard JSON clients (mcpServers)

Applies to ChatGPT Desktop, Claude Desktop, Cursor, Gemini CLI, Gemini antigravity, and the Gemini extension config.

{
  "mcpServers": {
    "duckduckgo": {
      "command": "uvx",
      "args": ["duckduckgo-mcp-server"]
    }
  }
}

Codex (~/.codex/config.toml)

[mcp_servers.duckduckgo]
command = "uvx"
args = ["duckduckgo-mcp-server"]

GitHub Copilot CLI (~/.copilot/mcp-config.json)

{
  "mcpServers": {
    "duckduckgo": {
      "type": "stdio",
      "command": "uvx",
      "args": ["duckduckgo-mcp-server"],
      "tools": ["*"]
    }
  }
}

OpenCode (~/.config/opencode/opencode.json)

{
  "mcp": {
    "duckduckgo": {
      "type": "local",
      "command": ["uvx", "duckduckgo-mcp-server"],
      "enabled": true
    }
  }
}

AITK (~/.aitk/mcp.json)

{
  "servers": {
    "duckduckgo": {
      "type": "stdio",
      "command": "uvx",
      "args": ["duckduckgo-mcp-server"]
    }
  }
}

Chrome DevTools setup (representative client snippets)

Use server key chrome-devtools through the managed MCPHub wrapper. The wrapper starts or reuses a dedicated visible Chrome on 127.0.0.1:9333 with profile ~/.cache/chrome-devtools-mcp-login, then attaches chrome-devtools-mcp with --browserUrl. This avoids the automation launch flags that chrome-devtools-mcp passes when it launches Chrome itself. Chrome opens lazily when the client invokes a browser-backed tool such as list_pages, navigate_page, or take_screenshot; restart the MCP client after config changes so it does not keep using an already-started server process.

Replace <repo> in the examples below with the absolute path to your local checkout.

Standard JSON clients (mcpServers)

Applies to ChatGPT Desktop, Claude Desktop, Cursor, Gemini CLI, Gemini antigravity, and the Gemini extension config.

{
  "mcpServers": {
    "chrome-devtools": {
      "command": "bash",
      "args": ["<repo>/scripts/mcphub/chrome-devtools-browser-url.sh"]
    }
  }
}

Codex (~/.codex/config.toml)

[mcp_servers.chrome-devtools]
command = "bash"
args = ["<repo>/scripts/mcphub/chrome-devtools-browser-url.sh"]

GitHub Copilot CLI (~/.copilot/mcp-config.json)

{
  "mcpServers": {
    "chrome-devtools": {
      "type": "stdio",
      "command": "bash",
      "args": ["<repo>/scripts/mcphub/chrome-devtools-browser-url.sh"],
      "tools": ["*"]
    }
  }
}

OpenCode (~/.config/opencode/opencode.json)

{
  "mcp": {
    "chrome-devtools": {
      "type": "local",
      "command": ["bash", "<repo>/scripts/mcphub/chrome-devtools-browser-url.sh"],
      "enabled": true
    }
  }
}

Crush (~/.config/crush/crush.json)

{
  "mcp": {
    "chrome-devtools": {
      "type": "stdio",
      "command": "bash",
      "args": ["<repo>/scripts/mcphub/chrome-devtools-browser-url.sh"]
    }
  }
}

---

Thinking and Reasoning (10)

Structured reasoning frameworks that guide agents through multi-step analysis, creative problem-solving, and decision evaluation.

sequential-thinking

Analysis Step-by-step reasoning with revision and branching. The reference implementation from Anthropic for dynamic, reflective thought processes.

atom-of-thoughts

Synthesis Decomposes complex problems into atomic, independent thought units that can be contracted and combined for robust answers.

shannon-thinking

Information Theory Information-theoretic reasoning inspired by Claude Shannon. Reduces noise and uncertainty through systematic problem decomposition.

structured-thinking

Memory Captures, retrieves, and revises thoughts with persistent history. Supports tagging and relevance-based retrieval across sessions.

cascade-thinking

Deep Dive Multi-stage reasoning cascade that escalates through progressively deeper analysis levels when simpler passes are insufficient.

crash

Verification Constraint-satisfaction reasoning with strict mode toggle. Enforces logical consistency checking on intermediate conclusions.

deep-lucid-3d

Exploration Three-dimensional problem analysis exploring depth, breadth, and lateral connections. Includes creative exploration mode.

lotus-wisdom-mcp

Perspective Eastern-philosophy-inspired reasoning combining multiple perspectives and reflective synthesis for balanced conclusions.

think-strategies

Routing Strategy-selection toolkit that matches reasoning patterns to problem types. Session management for ongoing analysis threads.

creative-thinking

Ideation Lateral and divergent thinking techniques for brainstorming, ideation, and breaking out of conventional solution spaces.

---

Search and Discovery (8)

Web search engines and content discovery services for retrieving up-to-date information from the internet.

brave-search

Web/Local Brave Search API with web, news, image, video, and local search. Includes AI-powered summarization of search results.

exa

Neural Search Neural search engine that understands meaning, not just keywords. Optimized for finding technical content, papers, and documentation.

g-search

Google Google Search results via SerpAPI. Familiar ranking and broad coverage across the public web.

duckduckgo

Privacy First Privacy-focused web search via DuckDuckGo. No tracking, no API key required.

ddgs

Metasearch Multi-engine search and news via DDGS (search_text, search_news). No API key. Optional full-page extraction is high-latency and token-heavy — use snippets by default.

arxiv

Academic arXiv paper search and metadata retrieval for research workflows, including direct access to paper IDs and summaries.

wikipedia

Encyclopedic Wikipedia content retrieval and lookup tools for encyclopedic background, entities, and topic overviews.

wayback

Archive Access archived pages from the Internet Archive Wayback Machine for historical snapshots and link recovery.

tavily

Research Tavily-powered web search, extraction, crawl, and deep-research workflows for current-event and multi-source investigations.

---

Web Fetching and Extraction (3)

Tools for retrieving, rendering, and extracting structured content from web pages and URLs.

fetch

HTTP Fetch Core URL fetcher from the MCP reference servers. Retrieves web pages and returns content in multiple formats.

fetcher

JS Rendering Browser-based URL fetcher with JavaScript rendering. Handles SPAs and dynamic content that simple HTTP fetches miss.

trafilatura

Content Extraction High-quality web content extraction using the Trafilatura library. Strips boilerplate to return clean article text and metadata.

scrapling

Opt-in Scraping CSS-targeted web scraping with stealth HTTP fetch and optional headless browser sessions. Opt-in via MCPHub research, web-read, live-browser, and related groups; not in default harness. Run scrapling install for browser tools.

---

Knowledge and Documentation (3)

Access library documentation, open-source wiki knowledge, and codebase analysis tools.

context7

API Docs Retrieves up-to-date documentation for libraries and frameworks directly from source. Eliminates hallucinated APIs and outdated references.

deepwiki

GitHub Wikis Queries the DeepWiki knowledge base of open-source project documentation. Ask questions about repos and read generated wiki pages.

repomix

Codebase Context Packs entire codebases into AI-friendly context. Supports local directories and remote repositories with grep and directory browsing.

---

Notebooks (1)

Live Jupyter notebook management and kernel execution. Requires a user-owned JupyterLab instance and opt-in MCPHub groups (notebooks, coding, heavy, etc.); not in default harness.

jupyter-mcp-server

Opt-in Notebooks Datalayer Jupyter MCP: list/read notebooks and cells, edit cells, execute code in kernels, multimodal outputs. Connect via MCPHub /mcp/notebooks or workflow groups; set JUPYTER_URL and JUPYTER_TOKEN in .env.mcphub.

---

Development Tools (3)

Browser automation, package management, and document processing for development workflows.

chrome-devtools

CDP Automation Chrome DevTools Protocol automation via Puppeteer. Navigate, click, fill forms, trace performance, inspect network requests, and take screenshots. Requires Node.js v22+.

package-version-check-mcp

Dependency Mgmt Checks latest stable versions across npm, PyPI, Go, Maven, Docker, Helm, Terraform, GitHub Actions, and mise-backed devtools via PyPI uvx.

docling

Doc Processing Converts documents (PDF, DOCX, PPTX, HTML) into structured Docling format. Create, edit, search, and export documents as Markdown.

---

Productivity (1)

Email, task management, and personal productivity integrations.

gmail

Inbox Zero Full Gmail access: search, read, send, draft, label, filter, and batch-manage emails. Supports attachment downloads.

---

Social (1)

Social media and professional network integrations.

linkedin

Professional Data LinkedIn profile scraping for professional networking data. Extract public profile information and connections.