Skip to content

feat: add Tabstack API tools for cloud-based extraction#329

Draft
lmorchard wants to merge 6 commits intodevelopfrom
tabstack-api-tools
Draft

feat: add Tabstack API tools for cloud-based extraction#329
lmorchard wants to merge 6 commits intodevelopfrom
tabstack-api-tools

Conversation

@lmorchard
Copy link
Collaborator

@lmorchard lmorchard commented Mar 6, 2026

Add three new tools that leverage the Tabstack API when a TABSTACK_API_KEY is configured:

  • tabstack_extract_markdown: extract clean markdown from web pages and PDFs (especially useful for PDFs which browsers can't read)
  • tabstack_extract_json: extract structured data per a JSON Schema
  • tabstack_generate_json: AI-transform page content via instructions

Tools are conditionally available (same pattern as search tools) and use the @tabstack/sdk TypeScript client.

Description

PR Type

  • New Feature

Checklist

  • I understand the code I am submitting
  • I have tested this code locally
  • New and existing tests pass locally (pnpm test)
  • I have added tests that prove my fix/feature works (if applicable)
  • Documentation was updated where necessary
  • I have read and followed the contribution guidelines

AI Usage

  • AI was used for drafting/refactoring

lmorchard and others added 6 commits March 5, 2026 15:15
Add three new tools that leverage the Tabstack API when a
TABSTACK_API_KEY is configured:

- tabstack_extract_markdown: extract clean markdown from web pages
  and PDFs (especially useful for PDFs which browsers can't read)
- tabstack_extract_json: extract structured data per a JSON Schema
- tabstack_generate_json: AI-transform page content via instructions

Tools are conditionally available (same pattern as search tools)
and use the @tabstack/sdk TypeScript client.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Add feature bullet, CLI option, and environment variable
documentation for the new tabstack_api_key config field.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Adds a fourth Tabstack tool that uses the /v1/agent/research endpoint
to search the web, analyze multiple sources, and synthesize answers
with citations. Supports fast and balanced modes.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Make the action loop prompt more opinionated about when to use
Tabstack tools over manual browsing. Key changes:
- Recommend tabstack_research FIRST for research-heavy tasks
- Explicitly state it's faster than manual search/navigate/extract
- Update PDF guidance to prefer tabstack_extract_markdown directly
- Frame all Tabstack tools as preferred alternatives, not just options

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The research endpoint returns flat objects with `message` as the
event type and `report`/`metadata` as fields on the final event,
not the `event`/`data` structure the SDK types suggest. Fixed both
the tool implementation and tests to match the actual API format.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
When a starting URL is given, the user wants the agent focused on
that specific site. Adjust prompt guidance so:
- tabstack_research is available but not the default first action
- webSearch is described as supplementary, not primary
- Both tools are still available if the site doesn't have enough info

This applies the same logic to both tabstack_research and webSearch.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant