Developer documentation

Model Context Protocol

Remote MCP exposes the same domain actions as REST over standard Streamable HTTP—no proprietary tool dialect required. Credentials, scopes, idempotency, and role checks still gate every tool call.

What MCP adds

MCP lets an authorized coding agent, IDE, cowork tool, or automation client discover AbuseZero tools and call them with structured input. MCP does not give a model extra authority; credentials, scopes, environment checks, role checks, idempotency, and audit rules still apply.

Remote endpoints

Connect Test clients to https://mcp.abusezero.com/mcp/test and Live clients to https://mcp.abusezero.com/mcp. A standards-compliant client needs only the endpoint, OAuth discovery with PKCE or an API-key bearer header, and standard MCP messages.

Transport independence

The server does not require OpenAI-specific or Anthropic-specific tool formats, sampling, elicitation, model APIs, or proprietary approval callbacks. The optional abusezero-mcp command included in @abusezero/sdk supports clients limited to local stdio.