autotel-devtools 27.1.0 → 27.2.1

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/NOTICE ADDED
@@ -0,0 +1,9 @@
1
+ autotel
2
+ Copyright 2026 Jag Reehal
3
+
4
+ This product is licensed under the Apache License, Version 2.0.
5
+ See the LICENSE file for the full license text.
6
+
7
+ "autotel" is a trademark of Jag Reehal. The Apache License, Version 2.0
8
+ does not grant rights to use the "autotel" name. See TRADEMARKS.md for
9
+ the trademark policy.
package/README.md CHANGED
@@ -177,6 +177,7 @@ process. Point them at the bound port, or free the original.
177
177
  - ✅ Resources view (derived from telemetry)
178
178
  - ✅ GenAI run summaries + narrated walkthrough
179
179
  - ✅ **Agents view**: observe coding agents (Claude Code, opencode), with cost split by model, effort, skill, sub-agent and prompt
180
+ - ✅ **WebMCP tools**: the full-page viewer offers its stored telemetry to an agent driving the page (see below)
180
181
  - ✅ Global time window shared by every tab
181
182
  - ✅ Live tail that freezes while you read, with a "N new" pill to catch up
182
183
  - ✅ Configurable telemetry limits (env vars)
@@ -328,6 +329,32 @@ The same store backs `autotel-mcp`'s `devtools` backend, so `autotel diagnose`,
328
329
  autotel diagnose errors --backend devtools --devtools-base-url http://localhost:4318
329
330
  ```
330
331
 
332
+ ### Asking an in-page agent
333
+
334
+ The full-page viewer also registers itself as a set of [WebMCP](https://github.com/webmachinelearning/webmcp)
335
+ tools, so an AI agent driving the browser can read the same telemetry without a
336
+ CLI, an API key or a screenshot. Open the viewer in a browser that supports
337
+ WebMCP and the tools are there:
338
+
339
+ | Tool | Answers |
340
+ | -------------------------- | ----------------------------------------------------------------- |
341
+ | `autotel_query_traces` | Search traces — one row each: name, service, status, duration |
342
+ | `autotel_get_trace` | Every span of one trace, by id |
343
+ | `autotel_list_errors` | What is failing, grouped by fingerprint |
344
+ | `autotel_query_logs` | Search log records, with the trace id when there is one |
345
+ | `autotel_webmcp_inventory` | The page's own WebMCP tool surface, including dropped annotations |
346
+
347
+ They take the [query language](#querying) above, so "show me failing checkouts
348
+ over 500ms" reaches the agent as `status = ERROR name contains checkout duration > 500`.
349
+ Results are projected down to the columns the panel lists and capped per call,
350
+ because a page of full span trees is mostly context the agent pays for and does
351
+ not read.
352
+
353
+ Read-only, and **full-page only**: the embedded widget is a guest in someone
354
+ else's page, where `document.modelContext` belongs to that page. Registered
355
+ against the browser's WebMCP API directly, with no runtime dependency; in a
356
+ browser without WebMCP nothing is registered.
357
+
331
358
  ## Configuration
332
359
 
333
360
  ### Environment Variables