Agent workflows

Why agents like webctx

Agents often need web context in a form that is easy to paste, store, diff, summarize, or cite. webctx returns terminal-friendly text instead of browser UI state.

The three useful primitives are:

search for candidate pages
read a page as markdown
map a site into URL candidates

Research loop

webctx search "OpenAI Apps SDK MCP tool annotations"
webctx read-link https://developers.openai.com/apps-sdk/reference

Use this when an agent needs recent docs or implementation context before editing code.

Documentation audit loop

webctx map-site https://docs.firecrawl.dev
webctx read-link https://docs.firecrawl.dev/introduction
webctx read-link https://docs.firecrawl.dev/api-reference/endpoint/scrape

Use this to gather the important pages before asking an agent to write or update docs.

Repo understanding loop

webctx read-link https://github.com/amxv/webctx
webctx read-link https://github.com/amxv/webctx/blob/main/CONTRIBUTORS.md
webctx read-link https://github.com/amxv/webctx/blob/main/docs/porting-status.md

GitHub README and blob URLs use the raw-content fast path, which avoids the scraping provider when the content is public.

Noise control

Use default and custom exclusions when web search is returning low-value result types:

webctx search "go cli release npm native binary" --exclude reddit.com,medium.com

Default exclusions already remove common video and social domains.

Keyword targeting

When a query needs a very specific phrase, use Exa keyword mode:

webctx search "firecrawl api" --keyword "maxAge excludeTags"

The keyword is truncated to five words before being sent as Exa include-text criteria.