MCP

Caelario tools inside your AI client

Model Context Protocol gives an MCP-compatible client Caelario's tool definitions and server instructions. The client chooses the appropriate tool from your natural-language request; the MCP server validates it, calls the same public API documented here, and returns normalized JSON.

1. Create a Caelario API key

Open Developers in the Caelario dashboard, create a workspace key, and copy the secret when it is shown. The guided installer asks for that key without placing it in the shell command or command history.

2. Install for Codex or Claude Code

Choose one command. The npm package starts a guided installer, prompts for the workspace key, and registers the user-wide STDIO server with the selected client. You do not edit TOML or JSON yourself.

npx -y caelario-mcp install codex

npx -y caelario-mcp install claude

1. 1
   Run one installer

   Node.js 20 or newer and the selected client CLI must be available on the machine.

2. 2
   Paste the key at the hidden prompt

   The package registers npx -y caelario-mcp serve and the API environment for that client.

3. 3
   Restart and verify

   Use /mcp, codex mcp list, or claude mcp list and confirm caelario is enabled.

Package and transport

caelario-mcp is the official local npm package. Codex and Claude Code launch it over STDIO; the package then calls https://api.caelario.com using the customer's workspace key. Run npx -y caelario-mcp serve only when testing with an MCP inspector—the client normally owns that process.

Available tools

The server exposes the complete live catalog across Instagram, TikTok, YouTube, X and Facebook. get_operation_guide explains the expected target, maximum and billing for any operation before work is created.

How the client chooses a workflow

1. 1
   Discover

   The client reads discover_platforms and get_operation_guide; it does not guess limits or prices.

2. 2
   Estimate

   estimate_credits shows the maximum reservation before a write.

3. 3
   Execute

   extract_now handles one bounded target. Larger or long-running work uses durable jobs or batches.

4. 4
   Wait and read

   wait_for_extraction can monitor a job; completed result tools support pagination for large JSON payloads.

Try it from any new task

Once enabled, Caelario is available to every new Codex task or Claude Code conversation on that machine; the conversation does not need prior project context.