Loading workspace…
How your tools get used
Customer workflows · 60 runs · Records tool cohort
B04 · Latest batch
- Tools provided
- 90% Confirmed in 54 of 60 runs
- MCP used
- 60% 36 of 60 runs called a tool
- Tools called
- 4 / 4 Across the whole batch
- Without MCP calls
- 24 Runs with no recorded MCP use
From availability to use
Two independent measurements over the same 60 runs. Their overlap is not shown.
Availability
54 runs confirm tool definitions were supplied. The remaining 6 have unknown availability.
Consideration
Supplying definitions does not show that an agent considered a tool. That decision is unobserved.
Recorded use
36 runs contain a tool call. The 24 without calls need prompt and response context to interpret.
Which tools get called
Runs can use more than one tool. Run counts below are not additive.
| Tool | Calls | Runs using tool | Run use rate | Returned errors |
|---|---|---|---|---|
| search_records | 30 | 27 / 60 | 45% | 0 |
| get_record | 30 | 25 / 60 | 41.7% | 3 |
| update_record | 15 | 7 / 60 | 11.7% | 9 |
| export_records | 9 | 9 / 60 | 15% | 0 |
Use across agents
Each illustrated agent ran the same prompts with two repeats.
| Agent | MCP use rate | Used / runs | Tool calls | Returned errors |
|---|---|---|---|---|
| General assistant | 65% | 13 / 20 | 31 | 4 |
| Reasoning assistant | 75% | 15 / 20 | 35 | 5 |
| Fast assistant | 40% | 8 / 20 | 18 | 3 |
Run cohorts are independent of production call volume. Tool invocation does not prove task success.