Problem Statement
Every search_code response starts with:
Top 8 results for: 'your query here'
The query is already present in Claude's context (it was just passed as an argument). Echoing it back in the output is redundant and wastes tokens on every single search call.
Proposed Solution
Remove the header line from the search_code output in server.py:
# Before
lines = [f"Top {len(results)} results for: '{query}'\n"]
# After
lines = []
Results are still clearly structured with [1], [2] prefixes and file:line (score) references, so removing the header does not reduce readability.
Alternatives Considered
- Shorten the header instead of removing it: saves fewer tokens and still redundant
- Keep it for human-readable logs: VecGrep output is consumed by Claude, not read by humans directly
Additional Context
- Zero quality risk
- Small but cumulative saving — every
search_code call across a session benefits
Problem Statement
Every
search_coderesponse starts with:The query is already present in Claude's context (it was just passed as an argument). Echoing it back in the output is redundant and wastes tokens on every single search call.
Proposed Solution
Remove the header line from the
search_codeoutput inserver.py:Results are still clearly structured with
[1],[2]prefixes andfile:line (score)references, so removing the header does not reduce readability.Alternatives Considered
Additional Context
search_codecall across a session benefits