Well, we try to... Check [?]openphacts.org[?], but that is a long way to go. Although, yes, even in the prototypes we do have KEGG (which might have to go because of copyright), ChemSpider (they are a partner) and some others.
The whole idea is to combine this kind of data in large datastores that know about things that identify the same object/concept and define fixed relationships (predicates) to other objects using semantic web approaches which can be queried using RDF. All that is part of what is usually called the semantic we, ad pharma tries to use it for relationships that involve chemicals and this potential drugs.
BTW the Manchester (Taverna) group is involved in that as well.
You mention ChemSpider in your question. ChemSpider is itself derived from many sources, including the other 3 that you mention (Pubmed, Pubchem and KEGG). And it comes with a set of web services. So I suspect that ChemSpider itself is one of the best toolkits there is.
It's unusual to find ready-made "connection toolkits" for multiple databases of any kind. In general, you write them yourself around the available APIs.
The Cactvs toolkit www.xemistry.com/academic for a free academic/educational version) supports most of these.
Advanced access to PubChem (both structures and assays, with full support for handling the native ASN.1 data, with lookup of CIDs, SIDs, names, etc, plus access of the PubChem structure set as a virtual multi-record file with query operations automatically forwarded to Entrez/PubChem Structure search)
Structure resolution from many identifiers via NCI Lookup service, KEGG or PubChem
Determination of structure IDs and display URLs for PubChem, KEGG, ChemSpider, and about 2 dozen other Internet structure sources
Some support for preconfigured Entrez/PubMed queries (names, MESH terms, etc.). more can easily be scripted.
Please contact me directly if you need help to get started.