Replies: 1 comment
-
|
Actually good idea agent use few tools calls and graph instead making multi calls Actually I created pr even to add support for mcp #1918 |
Beta Was this translation helpful? Give feedback.
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
-
Hey @decolua and the 9Router community,
9Router's whole promise is simple: never stop coding, stop bleeding tokens. RTK nails this after the agent decides to act — it compresses the
git diff,grep,ls,treeoutput before it ever reaches the model. Beautiful.But there's a second leak RTK can't reach from the proxy layer: the discovery phase. Before an agent edits a single line, it burns dozens of tool calls just figuring out where the relevant code lives — find, ls, grep, Read, repeat. On a large repo, that's where most of the budget quietly disappears.
That's the exact gap [CodeGraph](https://github.com/colbymchenry/codegraph) fills. It's a 100% local MCP server that pre-indexes your codebase into a queryable knowledge graph, so the agent asks the map instead of scanning the files.
Their published benchmarks (re-validated on Opus 4.8, June 2026):
Here's the part I like most: there's nothing to build. The two tools live at different layers and compose out of the box today —
RTK shrinks each trip. CodeGraph removes trips. The savings compound. Same audience, same mission, two different leaks plugged.
So the ask is small:
And for everyone reading — is anyone already running this combo? Curious what real numbers you're seeing. 👇
Links:
Beta Was this translation helpful? Give feedback.
All reactions