# Elite Reasoning MCP - Complete Protocol Specification

> Detailed JSON-RPC signatures, Pydantic type schemas, and execution protocol instructions for AI agents.

## Active Tools Reference

### `elite_prepare`
Start a non-trivial task. Retain run_id and obey the returned continuation after each later Elite call. Not the answer.

**Parameters:**
- `user_prompt`: String or typed parameter.
- `persist`: String or typed parameter.

### `elite_verify`
Run one check, persist evidence by run_id, then follow the returned continuation before answering.

**Parameters:**
- `check`: String or typed parameter.
- `query`: String or typed parameter.
- `draft`: String or typed parameter.
- `run_id`: String or typed parameter.
- `code`: String or typed parameter.
- `language`: String or typed parameter.
- `command`: String or typed parameter.
- `project_root`: String or typed parameter.
- `step_index`: String or typed parameter.
- `allowed_files`: String or typed parameter.
- `forbid_dependency_changes`: String or typed parameter.

### `elite_memory`
Search, record, or explicitly approve scoped memory items.

**Parameters:**
- `action`: String or typed parameter.
- `query`: String or typed parameter.
- `content`: String or typed parameter.
- `memory_type`: String or typed parameter.
- `scope`: String or typed parameter.
- `memory_id`: String or typed parameter.
- `trust_score`: String or typed parameter.
- `privacy_class`: String or typed parameter.
- `confirm`: String or typed parameter.

## Recommended Host Integration

For non-trivial coding tasks, hosts are asked (via IDE rules) to call `elite_prepare` first and then follow the returned `continuation` object:
```json
{"ServerName": "elite-reasoning-mcp", "ToolName": "elite_prepare", "Arguments": {"user_prompt": "<prompt>", "persist": true}}
```

MCP cannot force a host model to issue another call. Continuation directives, IDE rules, durable state, and evidence gates are layered mitigations, not absolute enforcement.