Current APIs, real usage, version diffs — compact, cited answers at a fraction of the tokens.
126,377 repos as continuously-fresh code graphs over MCP. And it's not just public
code: index your own repos and your
agents get the same grounded understanding of your system — connected to the
ecosystem it depends on.
agents: add the MCP endpoint · private repos: create a free workspace
indexing continuously — latest: github.com/google/snappy@1.3.1 · 43m ago
A free private workspace indexes your repos the same way: your agents can verify your APIs, trace calls across your repos, see the blast radius of a change before making it, and hop straight into the source of the open-source libraries you depend on — one connected graph, private to you, served over your own MCP endpoint.
the most-used graphs across the hub
> search("how do spans get pushed to the ingester") # grafana/tempo
1. Interface tempopb.PusherServer — modules/distributor · 4 implementers
2. Method Forwarder.ForwardTraces — modules/distributor/forwarder · "queues batches for the ingester"
3. Method distributor.PushTraces — called by 12 · calls 9
> callers("PushTraces")
receiver.Push → TracePushMiddleware.Push → distributor.PushTraces confidence 0.95$ claude mcp add --transport http codehub https://codehub.bot/mcp
search · symbol_context · callers · implementers · impact · load_source · datastores — over any graph on this page.
from_pretrained() of code graphs$ otcore fetch github.com/google/uuid@v1.6.0 $ otcore search uuid.db "parse a uuid from string"
7–50MB artifacts, offline-capable, hash-verified. sqlite, postgres and ladybug backends.
All your repos in one private graph — cross-repo calls and blast radius between your services, your own MCP URL, re-indexed on every push. GitHub and GitLab.
hosts the code — you still read it file by file.
the registry for models, not for code context.
generated prose about repos. Humans only.
the graph of every repo@version — computed facts, blast radius, symbol permalinks, and an MCP endpoint for agents.