Saussurea costus, a perennial herb belonging to the Asteraceae family, is a vital ingredient in traditional Chinese medicine. Increased demands for the herb have led to its widespread cultivation in China, but the corresponding increase in pesticide use has raised concerns about pesticide residues. Such residues would affect the safety and global market potential of Saussurea costus. Thus, a simple method is crucial to detect pesticide residues. The QuEChERS technique, in combination with gas chromatography-tandem mass spectrometry (GC-MS/MS), is commonly used for residue detection. However, traditional adsorbents may be unable to purify complex herbal mixtures well, affecting accuracy and instrument performance. Choosing suitable purification materials for Saussurea costus samples with complex matrices is of significant importance. This study focused on the detection of 35 prohibited pesticides in Saussurea costus. A rapid detection method was established by combining the QuEChERS technique with GC-MS/MS and utilizing a combination of multiwalled carbon nanotubes (MWCNTs), octadecylsilane-bonded silica gel (C18), and anhydrous magnesium sulfate (MgSO4) as the purification adsorbent. The samples were extracted with acetonitrile, purified by an improved QuEChERS process, subjected to GC-MS/MS analysis in multiple reaction monitoring (MRM) mode, and quantified using the internal standard method. The purification effects of four materials (C18, MWCNTs, N-propyl ethylenediamine (PSA), and graphitized carbon black (GCB)) and their optimal dosages were investigated by considering the matrix characteristics of the samples. An orthogonal experimental design was employed to optimize the ratio of adsorbent combinations, and the optimal adsorbent combination was determined to be 450 mg of MgSO4, 400 mg of C18, and 50 mg of MWCNTs. Matrix effect (ME) evaluation of the S. costus matrix showed that 31 target compounds strongly exhibited matrix-enhancement effects. Thus, matrix-matched calibration was employed in this study. Methodological investigation revealed that the standard curves for the 35 pesticides exhibited good linearity, with correlation coefficients (r2) greater than 0.9970. The average recoveries at three spiked levels ranged from 69.6% to 126.9%, and the relative standard deviations (RSDs) for parallel groups were all less than 10%. The limits of detection (LODs) and quantification (LOQs) ranged from 0.2 to 5.4 μg/kg and from 0.6 to 18.1 μg/kg, respectively. The developed method was used to screen and detect 35 pesticide residues in 20 batches of S. costus samples, and the target compounds were detected in six batches. The proposed method is simple, sensitive, and accurate. Thus, it is suitable for the rapid screening and detection of the 35 pesticide residues in S. costus and provides technical support for the cultivation, production, and quality control of the herb.