I would be great to be able to run the algorithms within graph-tool on data
from a graph database (such as neo4j) without having to have a static
GraphML representation of the data
Since Tinkerpop and pyBlueprints have their own graph data structure,
there is no other way than to convert it to graph-tool's format. There
are no plans to make the connection tighter, since this would involve
major rewrites.
However you can do better than writing it to a file by simply writing a
converter in Python. All you need to do is iterate through the edges and
vertices and add them to a new graph-tool graph, and vice versa.