@anvailabs/agentbrowser-mcp 1.7.0 → 1.8.0

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 CHANGED
@@ -17,18 +17,28 @@ input — are deliberately not exposed.
17
17
 
18
18
  | Tool | Purpose |
19
19
  |---|---|
20
- | `browser_create` | Create an isolated session (also creates the first page) |
20
+ | `browser_create` | Create an isolated session (also creates the first page; accepts seed cookies for authenticated re-entry) |
21
21
  | `browser_close` | Close a session, invalidating its refs |
22
22
  | `browser_navigate` | Navigate a page to an http(s) URL |
23
- | `browser_observe` | Semantic snapshot with stable element refs (`e<revision>_<ordinal>`) |
24
- | `browser_act` | click / fill / select / scroll / press — by ref, never by selector |
23
+ | `browser_observe` | Semantic observation with stable element refs (`e<revision>_<ordinal>`) |
24
+ | `browser_snapshot` | Self-contained page payload (url, title, revision, mode, `fields`) — one call to plan a whole form |
25
+ | `browser_plan` | Execute an ordered batch of fill/click/press/scroll steps in one call; stale refs self-heal, with a role+label guard under churn |
26
+ | `browser_act` | click / fill / select / scroll / press / dialogs — by ref, never by selector |
27
+ | `browser_extract` | Structured page extraction |
25
28
  | `browser_screenshot` | Capture optional visual evidence |
29
+ | `browser_pdf` | Page-to-PDF capture |
30
+ | `browser_cookies` | Export session cookies — persist and seed `browser_create` to re-enter an authenticated session |
31
+
32
+ The fast path for multi-field forms is `browser_snapshot` → `browser_plan`:
33
+ one call to read the page, one call to act on it — see the
34
+ [main README](../../README.md#the-fast-path-for-forms-snapshot-then-plan).
26
35
 
27
36
  ## Running it
28
37
 
29
38
  ### Release consumers: the single binary (recommended)
30
39
 
31
- Download `agentbrowser-mcp-<target>` from the [GitHub releases](../../releases)
40
+ Download `agentbrowser-mcp-<target>` from the
41
+ [GitHub releases](https://github.com/anvai-labs/agentbrowser/releases)
32
42
  (verify against `sha256sums.txt`), then point any MCP client at it — no Node,
33
43
  no pnpm, no repo checkout:
34
44
 
@@ -43,7 +53,7 @@ mcpServers:
43
53
 
44
54
  The binary is **only the MCP adapter** (TD-BROWSER-5 scope boundary): it needs
45
55
  a reachable AgentBrowser server over `AGENTBROWSER_BASE_URL` for anything
46
- beyond the protocol handshake. Deploy the server once (Docker image or
56
+ beyond the protocol handshake. Deploy the server once (release tarball, or
47
57
  `node packages/api/dist/bin.js`); every consumer process spawns the binary.
48
58
 
49
59
  ### Development: from the workspace
@@ -58,13 +68,14 @@ node packages/mcp-server/dist/bin.js
58
68
  | Variable | Meaning |
59
69
  |---|---|
60
70
  | `AGENTBROWSER_BASE_URL` | AgentBrowser server to proxy to (default `http://localhost:3000`) |
71
+ | `AGENTBROWSER_API_KEY` | Bearer key for the service, when it has `AGENTBROWSER_API_KEYS` configured |
61
72
  | `AGENTBROWSER_MCP_VERSION` | Override the reported `serverInfo.version` (debugging; the binary is otherwise stamped at build time from `package.json`) |
62
73
 
63
74
  ## Building and gating the binary
64
75
 
65
76
  ```sh
66
77
  pnpm --filter @anvailabs/agentbrowser-mcp compile # -> dist-bin/agentbrowser-mcp (needs bun)
67
- pnpm --filter @anvailabs/agentbrowser-mcp smoke # stdio handshake + 9-tool catalog gate
78
+ pnpm --filter @anvailabs/agentbrowser-mcp smoke # stdio handshake + full tool-catalog gate
68
79
  ```
69
80
 
70
81
  Cross-compile with `node scripts/compile.mjs --target=bun-linux-arm64 --outfile=...`.