interface RateLimitResult {
    allowed: boolean;
    reason?: string;
    retryAfterMs?: number;
    creditsUsed?: number;
    creditWarning?: string;
}
interface SessionStats {
    inFlight: number;
    creditsUsed: number;
    rpmCurrent: number;
}
type CostCheckResult = {
    allowed: true;
    warning?: string;
} | {
    allowed: false;
    warning: string;
};
export declare class McpRateLimiter {
    private sessions;
    private keyWindows;
    private sessionRpm;
    private keyRpm;
    private maxConcurrent;
    constructor(opts?: {
        sessionRpm?: number;
        keyRpm?: number;
        maxConcurrent?: number;
    });
    /**
     * Check all 4 dimensions before allowing a tool call.
     * Called at the /mcp endpoint level (SSE mode) or before tool execution (stdio mode).
     */
    check(sessionId: string, apiKey: string): RateLimitResult;
    /**
     * Called BEFORE a tool handler executes. Increments in-flight counter.
     */
    acquire(sessionId: string): void;
    /**
     * Called AFTER a tool handler completes (success or failure).
     * Decrements in-flight counter and records credits consumed.
     */
    release(sessionId: string, creditsConsumed: number): void;
    /**
     * Dimension 4: Cost-aware check. Called inside tool handlers after
     * fetching usage info from the backend. Returns a warning string
     * to append to tool responses, or null if credits are healthy.
     */
    checkCostThreshold(creditsRemaining: number, creditsTotal: number): CostCheckResult;
    /** Get session stats for health/debug endpoints */
    getSessionStats(sessionId: string): SessionStats | null;
    /** Cleanup when a session is destroyed */
    cleanup(sessionId: string): void;
    /** Cleanup when an API key has no more sessions */
    cleanupKey(apiKey: string): void;
}
export {};