Magnetic resonance imaging (MRI) has superior soft tissue contrast and lower interobserver variability compared to computed tomography and advances in equipment and pseudo-CT estimation have allowed for MR-only radiation therapy planning. Dedicated MR sequences have been used to localize paramagnetic structures with positive contrast, and most implanted seeds are gold fiducial markers (GFMs). We used a fast, dual-plane co-RASOR sequence to localize implanted GFMs with positive contrast in phantom and tissue to assess their resolution and registration accuracy of registration to CT. Off-resonant reconstructions of co-RASOR images were able to resolve GFMs down to 5 mm apart at 12 cm FOV. No systematic biases were observed by comparing registration of co-RASOR and bSSFP to CT images in an MR-compatible Lego phantom with a set of highly visible known points. The standard deviations of the MR to CT distance errors were <0.5 mm in all directions. We separated the component due to registration by comparing the two MR sequences, which had a maximum standard deviation of 0.36 mm in the SI-direction. Registration using the positive contrast points in a porcine sample phantom showed increased errors, but co-RASOR still performs acceptably with a target registration error of <0.75 mm. The dual-plane co-RASOR sequence could then be used for both registration and image tracking when performing MR-only radiation therapy planning.