Minimum-shift keying (MSK) limits problems associated with nonlinear distortion and is used in a variety of applications, like signal transmission from satellites and broadcasting. Although the optimal coherent MSK receiver simplifies to constant-complexity symbol-by-symbol detection, optimal blind reception of MSK takes the form of sequence detection (due to channel-induced memory) which has exponential (in the sequence length) complexity when implemented through an exhaustive search among all possible sequences. In this work, we present an algorithm that performs generalized-likelihood-ratio-test (GLRT) optimal blind sequence detection of MSK signals in flat fading with log-linear (in the sequence length) complexity. Moreover, for Rayleigh fading channels, the proposed algorithm is equivalent to the maximum-likelihood (ML) blind sequence detector. Simulation studies indicate that the optimal blind MSK detector attains coherent-detection performance when the sequence length is on the order of 100, offering a 5-6 dB gain over the typical single-symbol detector.