In creating such a system, the cost of the software and the expense of running the server(s) to host the data is microscopic compared to the expense of the army of knowledgeable people checking out references, cleaning up the barely legal (and semantically often complete nonsense) BibTeX entries offered by some journals, knit them up from scratch where not even the above is available.
Better create your central database, share it with your research group. Have everybody chip in by adding references to interesting papers found while browsing, whatever they cite in their publications/theses, and perhaps keep another file with publications generated locally (comes handy when asked for "publications of the group last year" or so). Publish guidelines, enforce them, perhaps occasionally organize a refer-thon to clean up entries and fill out missing details. If you ask everybody who reads a paper/document to add a short (3-5 lines) summary/abstract (they'll have to write one anyway for "state of the art"), a URL or other pointer where to find the document, you create an (at least locally) very valuable resource.
If it grows enough, or as a side product of the work, publish it on the webpage of the group. Might even ask for contributions from outside while you are at it.