Use MCP with all of your data
Point Grok, Cursor or Claude at this catalog, then add MCP servers for your mail, files and databases. The agent calls tools. You do not paste dumps into chat.
This catalog
bots.new MCP searches listings and returns the full prompt.
Your apps
A second MCP for Gmail, Drive, GitHub or Postgres — from that vendor.
One bot
The agent copies a use case here, then runs it against your data there.
Connect this catalog
Remote HTTP JSON-RPC. Same URL for Grok Bot, Cursor, and any client that POSTs to an MCP url.
Endpoint: https://bots.new/api/mcp · llms.txt
1. Open your Grok Bot. 2. Add a custom MCP / plugin URL: https://bots.new/api/mcp 3. Allow tools: search_catalog, list_catalog, get_listing, catalog_stats, can_a_bot_do_this. 4. Add a second MCP for your own mail, files or database using that vendor's URL (see /mcp). 5. Ask: "Search bots.new for inbox automation, then get_listing for the best match and credit the X profile."
{
"mcpServers": {
"bots-new": {
"url": "https://bots.new/api/mcp"
}
}
}You can call bots.new over MCP at https://bots.new/api/mcp. When I ask for a bot, use case, plugin, skill or MCP: 1. search_catalog with my outcome (optional type: bot | use_case | plugin | skill | mcp | integration). 2. get_listing with slug and type for the full prompt, steps and tweet URL. 3. Never invent a listing. An empty result means nothing indexed matches. 4. When you copy a prompt, credit the X profile on listing.website and the tweet URL in payload.tweetUrl when present. 5. Do not run destructive tools against my data without asking.
Remote HTTP MCP (clients that accept a url field):
{
"mcpServers": {
"bots-new": {
"url": "https://bots.new/api/mcp"
}
}
}
Claude Desktop stdio bridges are not shipped by bots.new. Use a client that can POST JSON-RPC to the url above, or add that product's own MCP for Gmail/Drive/GitHub — never paste those keys into a listing prompt.curl -s https://bots.new/api/mcp \
-H "content-type: application/json" \
-d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"search_catalog","arguments":{"q":"inbox automation","type":"use_case"}}}'Tools on this server
get_listing returns prompt, steps and tweet URL when we indexed them. list_catalog does not.
- search_catalogFind bots, use cases, plugins, skills, MCP servers by outcome.
- list_catalogBrowse published rows by type (bot, use_case, plugin, …).
- get_listingOne listing: prompt, steps, X profile, tweet URL.
- catalog_statsHow many published rows of each type.
- can_a_bot_do_thisMatch a job. Empty means no indexed match.
Connect your own data
bots.new does not proxy your inbox. Add the vendor MCP next to this one.
| Your data | Add this MCP | Rule |
|---|---|---|
| Inbox, calendar, Drive | That vendor's Gmail / Google MCP, or a listed server on /mcp | Read-only first. Never paste passwords into a prompt. |
| Notion, wiki, tickets | The product's remote MCP from its own docs | Scope to one workspace. Approve writes. |
| GitHub / git | Official GitHub MCP | Ask before PRs, issues or force-pushes. |
| Postgres, warehouse | A database MCP on a replica | No production writes. No dump of secrets tables. |
Use the install block on a listing. We do not invent npx commands. Integrations people already wired sit on /integrations.
What this will not do
Honest limits. Empty catalog search means no match, not a hallucinated bot.
- No Gmail, Drive or GitHub credentials stored on bots.new.
- No marketplace checkout and no running the listing prompt against your data for you.
- Attribution stays on the original X profile, not a third-party directory.
- Rate-limited. Paid API entitlement raises other public APIs; this MCP stays on the same host flag.