book_flight
Last 7 days
Calls
Tool calls in this window that reached a tool this server has, answered or failed. Calls to tools it lacks, resource reads and prompt gets are counted in their own cards.372
Nothing to compare with in the 7 days before
Error rate
Failed calls out of all calls: a result the tool marked as an error, an exception it threw, or arguments refused before it ran.22%
Elevated82 failed calls
Response time
How long your server took to answer, measured by the SDK inside it, so the network to the client is not included. The median is the typical call; 19 in 20 calls finish within the 95th percentile.976ms
median, 2.8s at the 95th
Called from
The client with the most calls, by the name it gives itself when it connects. A client can call itself anything, so the name is as reported.Claude Code
31% of calls, 4 other clients
Calls over time
Show the numbers
| Time | Calls | Errors |
|---|---|---|
| 8 | 2 | |
| 9 | 6 | |
| 4 | 1 | |
| 4 | 0 | |
| 31 | 13 | |
| 20 | 9 | |
| 25 | 11 | |
| 4 | 2 | |
| 12 | 2 | |
| 19 | 8 | |
| 26 | 7 | |
| 8 | 4 | |
| 19 | 8 | |
| 31 | 5 | |
| 8 | 0 | |
| 6 | 0 | |
| 19 | 0 | |
| 17 | 1 | |
| 14 | 1 | |
| 5 | 1 | |
| 15 | 0 | |
| 27 | 0 | |
| 18 | 0 | |
| 7 | 0 | |
| 3 | 0 | |
| 6 | 1 | |
| 6 | 0 | |
| 0 | 0 | |
| 1 | 0 |
Versions
This tool's calls by the server version that answered them
| Version | First seenWhen mcpspan first saw a call answered by this version, in the data still kept. | Calls | ErrorsThe share of calls that failed: an error the tool reported, an exception it threw, or arguments the server refused. Calls to names the server does not have are counted apart. | MedianHow long calls took: half of them were faster than the median. p95 under it: 95 in 100 were faster, so it shows the slow tail the median hides. |
|---|---|---|---|---|
| 1.4.2 | 13 | 7.7% | 1.0sp95 1.6s | |
| 1.4.1 | 80 | 2.5% | 905msp95 2.0s | |
| 1.4.0 | 167 | 20% | 1.3sp95 3.0s | |
| 1.3.0 | 112 | 41% | 597msp95 1.9s |
Response times
Calls by how long they took
Show the numbers
| Response time | Calls | Share |
|---|---|---|
| Up to 1ms | 5 | 1.3% |
| 1ms to 2ms | 37 | 9.9% |
| 2ms to 3ms | 19 | 5.1% |
| 3ms to 5ms | 11 | 3.0% |
| 5ms to 8ms | 0 | 0.0% |
| 8ms to 12ms | 0 | 0.0% |
| 12ms to 20ms | 0 | 0.0% |
| 20ms to 30ms | 0 | 0.0% |
| 30ms to 45ms | 0 | 0.0% |
| 45ms to 65ms | 0 | 0.0% |
| 65ms to 100ms | 0 | 0.0% |
| 100ms to 150ms | 0 | 0.0% |
| 150ms to 250ms | 0 | 0.0% |
| 250ms to 400ms | 1 | 0.3% |
| 400ms to 600ms | 27 | 7.3% |
| 600ms to 1s | 91 | 24.5% |
| 1s to 1.5s | 89 | 23.9% |
| 1.5s to 2.5s | 68 | 18.3% |
| 2.5s to 4s | 21 | 5.6% |
| 4s to 6s | 3 | 0.8% |
| 6s to 10s | 0 | 0.0% |
Right before it went wrong
Before 82 repeated or failed calls, in their sessions
What the agent called just before book_flight repeated itself or failed. One call standing out usually did not do what the agent expected: a click that did not land, a search that came back empty.
| Called before | Client | Repeated | Failed |
|---|---|---|---|
| book_flight | Claude Code | 0 | 13 |
| seat_map | Claude Code | 0 | 9 |
| seat_map | Claude | 0 | 8 |
| book_flight | Other | 0 | 7 |
| seat_map | ChatGPT | 0 | 7 |
| seat_map | Cursor | 0 | 7 |
| search_flights | ChatGPT | 0 | 6 |
| book_flight | ChatGPT | 0 | 5 |
| book_flight | Claude | 0 | 5 |
| book_flight | Cursor | 0 | 4 |
| book_flight | MCP Inspector | 0 | 3 |
| search_flights | Claude | 0 | 2 |
| search_flights | MCP Inspector | 0 | 2 |
| search_flights | Other | 0 | 2 |
| search_flights | Claude Code | 0 | 1 |
| seat_map | Other | 0 | 1 |
In the tool list
What its definition costs every session, and how many of those sessions used it
- DefinitionWhat the tool's name, title, description and input schema take in the model's context, read at the start of every session whether the tool is called or not. Estimated at four bytes a token; the exact count depends on the model.
- ≈ 605 tokens, 2.4 kB
- Share of the list
- 23%
- Shown inSessions whose client was sent this tool in the list of tools, in this window.
- 912 sessions
- Used inOf the sessions shown the tool, how many called it at least once. A tool shown often and used rarely may need a clearer description, or may not be needed.
- 227 (25%)
Response size
Over 290 answers; what the client's context takes in
- Typical (median)Half of the answers were smaller than this, as compact JSON. The size is all that is measured; the content never leaves the server.
- 1.3 kB
- Large (95th percentile)95 in 100 answers were smaller. A large one fills the agent's context, and every later step pays for it.
- 2.1 kB
- Largest
- 2.2 kB
How it fails
82 failed
- Invalid arguments
- 72
- Crashed
- 10
What it said, most often
Seat 14C was taken while you were booking
4 times, crashed, last
Upstream booking service timed out after 30s
3 times, crashed, last
Card was declined by the issuer
3 times, crashed, last
Called from
- Claude Code
- 116 (31%)
- Claude
- 81 (22%)
- ChatGPT
- 67 (18%)
- Cursor
- 59 (16%)
- Other
- 31 (8%)
Which arguments were invalid
72 calls refused before the tool ran
- passengers
- 70 calls
- flight
- 1 call
- payment
- 1 call
The argument agents get wrong most is usually one whose description leaves room for the wrong shape.
Parameters agents send
Names and types only
A name the schema does not have, or one that arrives as more than one type, is usually a description the agent is misreading.
| Name | Sent as | In calls |
|---|---|---|
| flight | string | 100% |
| passengers | array, number | 100% |
| payment | object | 100% |