MCP server
Six tools your agent can call.
VibeHunt exposes a Model Context Protocol endpoint over Streamable HTTP. Any compatible client — Claude Desktop, Claude Code, Cursor, Cline, Windsurf — can connect with a single line of config and gain the ability to search the registry, fetch manifests, and draft new ones for software the user just built.
No auth, no API key, no rate limit. The protocol forbids accounts; the registry is public.
The six tools
What an agent gains the moment it connects.
vibehunt_searchFind software matching a described need. Returns ranked Capabilities with score breakdown.
- intent*
- Plain-text description of what the user wants.
- category
- Optional category ID (see vibehunt_list_categories).
- constraints
- Hard requirements as core: term IDs (e.g. 'core:local-first').
- limit
- Max results, 1–20. Defaults to 5.
vibehunt_get_capabilityFetch the full Capability manifest for a given ID or slug.
- id_or_slug*
- Either the full ID (cap:example:logseq) or slug (logseq).
vibehunt_list_categoriesList the 15 standard app-store-shaped categories.
vibehunt_list_constraintsList the 20 core: vocabulary terms with plain-English labels and descriptions.
vibehunt_draft_manifestGenerate a publishable Capability manifest plus the .well-known URL the publisher must host it at to verify the domain.
- name*
- Software name.
- homepage*
- Canonical homepage URL — the domain becomes the verification anchor.
- category*
- Category ID from vibehunt_list_categories.
- intent*
- One sentence about the problem the software solves.
- constraintsSatisfied
- core: terms the software guarantees.
- maturity
- Defaults to 'experimental'.
- useDidWeb
- Defaults to true — ties the manifest to the homepage's domain.
vibehunt_submit_capabilitySubmit a hosted Capability manifest URL to the registry. Server fetches it, validates the schema, runs did:web verification, and indexes it live on success — searchable and browsable on the next request, no queue. Re-submission updates the entry.
- url*
- Publicly-fetchable URL of the hosted manifest — typically https://DOMAIN/.well-known/vibehunt/[slug].json.
Connect your agent
Pick your client. Each is a one-time config edit.
Claude Code
claude mcp add --transport http vibehunt https://vibehunt.tech/api/mcpRun this once in your terminal. The server appears in /mcp inside any Claude Code session.
Claude Desktop
~/Library/Application Support/Claude/claude_desktop_config.json (macOS) %APPDATA%\Claude\claude_desktop_config.json (Windows)
{
"mcpServers": {
"vibehunt": {
"url": "https://vibehunt.tech/api/mcp"
}
}
}Restart Claude Desktop after editing. Older versions that don't support remote URLs can use the mcp-remote proxy: replace the value with { "command": "npx", "args": ["-y", "mcp-remote", "https://vibehunt.tech/api/mcp"] }.
Cursor
~/.cursor/mcp.json (or Settings → Tools → MCP)
{
"mcpServers": {
"vibehunt": {
"url": "https://vibehunt.tech/api/mcp"
}
}
}Cline
VS Code: Cline panel → MCP Servers → Configure → cline_mcp_settings.json
{
"mcpServers": {
"vibehunt": {
"url": "https://vibehunt.tech/api/mcp",
"transportType": "streamable-http"
}
}
}Windsurf
~/.codeium/windsurf/mcp_config.json
{
"mcpServers": {
"vibehunt": {
"serverUrl": "https://vibehunt.tech/api/mcp"
}
}
}Any other MCP-compatible client works too — point it at the Streamable HTTP endpoint https://vibehunt.tech/api/mcp with no authentication.
What to ask your agent
Example prompts that exercise the tools.
Why an agent-shaped registry
Most software discovery surfaces — app stores, search results, social feeds — assume a human is the one doing the looking. Increasingly, they aren't. An agent that just helped a user build a feature can search the registry for prior art before reinventing it; an agent that just helped publish a tool can produce a verifiable manifest the moment the user is ready to ship. The MCP server is the publish-side tool that closes the loop between vibe-coded creation and discoverability.