MCP

Connect Claude, Cursor, VS Code and other MCP clients to the public read-only Incantory MCP server at https://incantory.ai/mcp.

Incantory runs a public, read-only Model Context Protocol server. Connect it to your AI client and the model can search the registry, read prompts, render them with variables and convert them to your coding tool's format.

https://incantory.ai/mcp

Transport: streamable HTTP.
Anonymous access works, rate-limited per IP.
With a token (an API token with the read scope, sent as Authorization: Bearer inc_...), limits are higher, your private and unlisted items become readable, and your starred and collected prompts appear as MCP prompts.

Tools

All tools are read-only.

Tool	What it does
`incantory_search`	Search prompts, skills, rules and agents (hybrid keyword and semantic search), or list by kind, tag and model
`incantory_get_prompt`	Get a prompt by ref (`owner/slug`, `owner/slug@v3`, `owner/slug@production`) with its labels, variables and entry file
`incantory_list_versions`	A prompt's versions, newest first, with labels and yanked marks
`incantory_render`	Fill in a prompt's variables and return the text or chat messages; nothing is sent to a model
`incantory_convert`	Convert a prompt to a coding-tool format (Claude skill or agent, Cursor `.mdc`, Copilot instructions, `AGENTS.md`, …) and return the files to write
`incantory_get_examples`	Approved example outputs for a prompt, by model or version
`incantory_related`	Related prompts: similar content, shared tags, forks and remixes

Prompts as slash commands

The server also exposes MCP prompts (prompts/list, prompts/get). With a token, the list is your own shortlist: the prompts in your collections, and your starred prompts. Anonymously it is the current top public prompts. Each appears in the client as a command named owner.slug, with the prompt's variables as its arguments.

Connect a client

Each block below is complete; copy it into the file named in the heading. To send a token, add the Authorization header shown in the last section.

Claude Code

bash

claude mcp add --transport http incantory https://incantory.ai/mcp

With a token:

bash

claude mcp add --transport http incantory https://incantory.ai/mcp --header "Authorization: Bearer inc_..."

Claude Desktop

In Claude Desktop you can add a remote server under Settings → Connectors, using the URL above. To configure it in claude_desktop_config.json instead, bridge it with mcp-remote:

json

{
  "mcpServers": {
    "incantory": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://incantory.ai/mcp"]
    }
  }
}

Cursor

.cursor/mcp.json in your project, or ~/.cursor/mcp.json for every project:

json

{
  "mcpServers": {
    "incantory": {
      "url": "https://incantory.ai/mcp"
    }
  }
}

VS Code

.vscode/mcp.json in your workspace (VS Code uses a top-level servers key):

json

{
  "servers": {
    "incantory": {
      "type": "http",
      "url": "https://incantory.ai/mcp"
    }
  }
}

Other clients

Any client that speaks streamable HTTP can use the URL directly. The common mcpServers shape:

json

{
  "mcpServers": {
    "incantory": {
      "type": "http",
      "url": "https://incantory.ai/mcp",
      "headers": { "Authorization": "Bearer inc_..." }
    }
  }
}

Create a read token at Settings → Security. Leave the headers entry out to connect anonymously.

The private MCP server

Incantory also runs a private MCP server for site administration: moderation, ingest, users, catalog and the other admin workflows, with the full read and write tool set. It is for Incantory admins only and requires an API token with the admin:mcp scope, which only admin accounts can create. It is not publicly reachable.