← MCP UI docs · Methodology · canonical URL
Executor.sh vs ClawQL — executor-cmp-001 (live)
Same task · vercel/next.js pulls.list · cl100k_base · Layer 1 tool defs + Layer 2 tool result · focus=input by default
Layer 1 (tool definitions): Right column uses 115 tok from our live Executor MCP install (execute only). executor.sh publishes ~1,044 tok for their codemode chart — a different measurement. ClawQL measured 394 tok (search+execute codemode). 394 is smaller than the published Executor figure, not smaller than live execute-only 115.
Layer 2 (tool result): Layer 2 is 100% of Executor's input and 70% of ClawQL's — almost the entire Executor bill is uncacheable tool payload after pulls.list.
Matched conditions: cl100k_base · focus=input (default) · executor-cmp-001 live · vercel/next.js
Combined (L1+L2, input only): Executor 143,581 vs ClawQL 1,301 = 110× (live Executor L1 115). With Executor's published homepage L1 (~1,044), combined would be 111× (144,510 vs 1,301). Layer 2 alone: 158× (143,466 vs 907). Model output omitted — same discipline as the built-in compressed vs fat demos.
Harness prompt
Vault / system-seed
Tool schema
User
Tool result
Agent reasoning
Other
ClawQL (search + execute + fields projection)
L1 394 + L2 907 = 1,301 tok input · 1,301 input · tool result 70% of input
| Source | Tokens |
|---|
| Tool schema |
394 |
| Tool result |
907 |
Executor (live MCP install + full REST list)
L1 115 live (homepage ~1,044) + L2 143,466 = 143,581 tok · 143,581 input · tool result 100% of input
| Source | Tokens |
|---|
| Tool schema |
115 |
| Tool result |
143,466 |
Source: executor-cmp-001.live.json · Canonical (blog links): /mcp-ui/trace/compare/executor · JSON: compare
· Sessions: demo-executor-cmp-clawql
· demo-executor-cmp-executor
· Generic compare: compressed vs fat · include outputs (not used in blog headline)