Crabtalk Hub

A community registry for AI agent resources — MCP servers, skills, agents, and commands.

Resources are indexed as scope/config.toml files in this repository. Anyone can add resources by opening a PR.

Index Format

Each resource lives at <scope>/<config>.toml:

microsoft/playwright.toml
notion/notion.toml
my-org/my-tool.toml

scope is typically the author, organization, or tool namespace. One .toml file can declare multiple related resources under the same scope.

Manifest Structure

Every manifest starts with a [package] section, followed by one or more resource sections.

[package]
name = "my-package"
description = "Short description for hub display"
logo = "https://example.com/logo.png"
repository = "https://github.com/my-org/my-package"
keywords = ["tag1", "tag2"]

Field	Required	Description
name	yes	Package name
description	no	Short description for hub display
logo	no	Logo URL for hub display
repository	no	Source repository URL (required for skills)
keywords	no	Searchable tags for hub discovery

Resource Sections

[mcps.*] — MCP Servers

Registers an MCP server. Merged into the user's crab.toml on install.

[mcps.notion]
command = "npx"
args = ["-y", "@notionhq/notion-mcp-server"]
env = { NOTION_TOKEN = "" }

Field	Required	Default	Description
command	yes*	—	Executable to spawn (stdio transport)
args	no	[]	Command-line arguments
env	no	{}	Environment variables
name	no	map key	Display name
auto_restart	no	true	Auto-restart on failure
url	no	—	HTTP URL for streamable HTTP transport (replaces stdio)

*Either command or url should be set.

[skills.*] — Skills

Downloadable skill directories, copied from package.repository to ~/.crabtalk/skills/<key>/ on install.

[skills.playwright-cli]
description = "CLI for common Playwright actions."
path = "skills/playwright-cli"

Field	Required	Description
description	yes	Skill description
path	yes	Path within the source repo to the skill directory
name	no	Display name (defaults to map key)

[agents.*] — Agents

System prompt bundles that can declare skill dependencies. Prompt files are copied to ~/.crabtalk/agents/<key>.md on install.

[agents.my-agent]
description = "An agent with a bundled prompt."
prompt = "prompts/my-agent.md"
skills = ["playwright-cli"]

Field	Required	Description
description	yes	Agent description
prompt	yes	Path to prompt .md file (relative to scope dir)
skills	no	Skill keys from the same manifest to auto-install

[commands.*] — Commands

Metadata for locally-installed command binaries. Hub registers them in crab.toml but does not download or manage the binary itself.

[commands.search]
description = "Meta-search aggregator (DuckDuckGo, Wikipedia)"
binary = "crabtalk-search"
subcommand = "mcp"

Field	Required	Description
description	yes	Human-readable description
binary	yes	Executable name (resolved via PATH) or absolute path
subcommand	yes	Clap subcommand for service management

Installed commands are managed via crabtalk <command> start|stop|run|logs.

Examples

• microsoft/playwright.toml — MCP server and CLI skill for browser automation.
• notion/notion.toml — Notion MCP server for pages, databases, and blocks.
• vercel/agent-browser.toml — Browser automation CLI for AI agents.
• crabtalk/search.toml — Meta-search aggregator command.
• crabtalk/telegram.toml — Telegram bot gateway command.

Add a Resource

1. Fork this repo
2. Create <scope>/<config>.toml following the format above
3. Open a PR — the PR template will guide you through the checklist

License

MIT