Coming soon

MCP server

Give AI clients fast, cheap decision tools through the Model Context Protocol.

This capability is not live yet. The reference below describes how it works at launch. Join the waitlist to get Early Access.

The MCP server exposes ClassifierHub decisions as tools. It calls the same API as your code, with your API key, so credits, limits and results are identical. It holds no model credentials of its own.

Remote server

{
  "mcpServers": {
    "classifierhub": {
      "url": "https://classifierhub.com/api/mcp",
      "headers": { "Authorization": "Bearer ch_live_..." }
    }
  }
}

Local server (stdio)

{
  "mcpServers": {
    "classifierhub": {
      "command": "npx",
      "args": ["-y", "@classifierhub/mcp"],
      "env": { "CLASSIFIERHUB_API_KEY": "ch_live_..." }
    }
  }
}

Tools

Tool	What it does
`classifierhub_decide`	Run a template, saved decision or yes/no question.
`classifierhub_classify`	Pick one label from options you provide.
`classifierhub_score`	Place the input on an ordered scale.
`classifierhub_extract`	Field extraction. Coming soon: returns FEATURE_DISABLED.
`classifierhub_batch`	Run a decision over many inputs and wait for results.
`classifierhub_templates`	List built-in templates.

Example: agent guardrail

Ask your agent to call classifierhub_decide with template: "agent-action-guardrail" before running a tool that changes data, and to stop or ask you when the verdict is review or reject.