The viscosity of micro liquid can be quickly measured based on the construction of the fundamental relationship between the capillary guided wave attenuation rate and the liquid viscosity within the waveguide. It is significant for industrial and medical testing scenarios in which the sample is scarce. However, it is necessary to reduce the pipe size to the capillary level ( the outer diameter is 2 mm or less) to achieve micro-upgrade sample measurement. Traditional guided wave excitation methods face challenges in achieving non-contact transducer assembly and pure guided wave excitation in capillaries. In this paper, a capillary longitudinal guided wave detection sensor suitable for the viscosity detection of trace liquid is developed based on the principle of magnetostriction. Pure L(0,1) modal waveguide are successfully excited on a capillary with an outer diameter of 1. 4 mm and a wall thickness of 0. 1 mm. The sample volume is only 113 μL is required for a single measurement, with good repeatability. The feasibility and practicality of the design are proved on the basis of simulation and experimental exploration of the relevant influencing factors including the distance between the permanent magnet and capillary, capillary wall thickness and excitation frequencies. The results show that the optimal distances between the permanent magnets and capillaries differ at the transmitting and receiving ends, which are 10 mm and 7 mm respectively for the maximum signal amplitude. In the frequency range with low dispersion (below 500 kHz), the detection sensitivity always increases with the frequency. In addition, reducing the wall thickness thinning of capillaries can also improve the viscosity detection sensitivity. Finally, a comparison of the capillary guided wave method with the cone-plate method for measuring standard viscosity liquids showed an error range of no more than 3. 04% , which verifies that the microfine tube guided wave method can achieve high-precision measurement of the viscosity of micro-volume liquid.