You may want to take a look at structural classification databases such as CATH and SCOP. Both should provide downloadable files of their data that provide mappings between pdb ids, domains and their classification. At least CATH has free text search (e.g. 'TIM barrel') that returns pdbs.

There are very good search/retrieve facilities at the PDB website itself.

For example:

• Start typing "TIM" in the search box and the site auto-suggests matches, including "TIM barrel" (2783 structures under "Structural Domains")
• Refine your query from the results page
• Create a report for download in CSV format, which includes the PDB IDs
• Use the download page or the RESTful services to retrieve structures using IDs

So I suggest you spend some time getting to know the tools at the PDB website; often, there is no need for 3rd party services or programming.

doing same thing that chris recommended, but with Astral set would work as well. Astral is a subset of SCOP, but has non-redundant proteins (< 30% sequence identity).