@gaiventures/pilotfish-mcp 0.1.5 → 0.1.6
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +4 -2
- package/cli.js +80 -80
- package/package.json +1 -1
package/README.md
CHANGED
|
@@ -89,7 +89,7 @@ loopback server answers only loopback `Host` headers (DNS rebinding). Set `PILOT
|
|
|
89
89
|
| `browser_select` `{ref, values}` | Pick `<select>` option(s) by label or value. |
|
|
90
90
|
| `browser_press` `{key}` | A key or shortcut (`Enter`, `Escape`, `Control+a`). |
|
|
91
91
|
| `browser_scroll` `{direction}` | Scroll about one screen; returns the screen it shows. |
|
|
92
|
-
| `browser_actions` `{actions}` | Up to 10 steps in one call
|
|
92
|
+
| `browser_actions` `{actions}` | Up to 10 steps in one call, each `{action, …}`: `click`, `select` and `type` take a `ref`; `type` takes `text` (and `submit`), `select` `values`, `press` `key`, `scroll` `direction`, `wait` `seconds`. E.g. a form's fields then submit; each goes through the same checks as its own tool, and a step missing a field, or with another action's, is refused before any step runs. Stops at the first that fails or after one that moves the page to another address. |
|
|
93
93
|
| `browser_query` `{expression}` | Read just the values you need with a read-only JavaScript expression (JSON, at most 4,000 characters): a results list, prices, counts. It runs in a world of its own where clicks, typing, submits and network writes are refused. The cheapest way to read. |
|
|
94
94
|
| `browser_find` `{query}` | Refs of the elements on the whole page that match the words, best first; partial matches are marked. Several targets can be separated by commas. |
|
|
95
95
|
| `browser_get_text` `{max_chars?, selector?}` | Readable text of the whole page (PDFs too), or of the first element matching `selector` (`main`, `#results`), secrets masked, up to `max_chars` (default 20,000). |
|
|
@@ -104,7 +104,9 @@ one it re-reads (a prompt-cache write against a read). So an action on a page th
|
|
|
104
104
|
(at most 8) returns only the lines that changed (`CHANGES ONLY since the page snapshot N browser results back`), each
|
|
105
105
|
under its unchanged parent line marked `~ `, with what is gone listed by ref. Refs are stable across snapshots, so the
|
|
106
106
|
lines not repeated still work; a change of more than half of the screen gets the whole screen, and
|
|
107
|
-
`browser_snapshot` always does.
|
|
107
|
+
`browser_snapshot` always does. The same goes for what the agent reads before any page: each tool's use is said once,
|
|
108
|
+
in its own description, and the server instructions keep only the two rules that span the tools (a "cheapest",
|
|
109
|
+
"latest" or "most" is the site's own sort; what to do on `NEEDS A PERSON`).
|
|
108
110
|
|
|
109
111
|
**Pages only a person can pass.** After every acting tool the page is classified. A sign-in with no credentials, a
|
|
110
112
|
one-time code or 2FA prompt, a device approval, a sign-in link, a passkey, 3-D Secure or a CAPTCHA adds a
|