update to v2 and streamable http

[tomkosm]

Streamable http from #63

[rafaelsideguide]

• screenshot@fullPage doesn't exist anymore. use { type: "screenshot", fullPage, quality, viewport }
• remove extract
• jsonOptions was removed. must send options w format: { type: "json", prompt, schema }

[rafaelsideguide]

you can just remove this

[rafaelsideguide]

• same changes we did in scrape for scrapeOptions.formats in crawl.

• for "asynchronous crawl" we're now using "start crawl" or "start crawl job", I think it's easier for the agent to understand

• there's a "client.asyncCrawlUrl(url, { ...options, origin: 'mcp-server' })" -> is now "startCrawl"

[rafaelsideguide]

hey @tomkosm, checked the tools and found a few points that need attention:

• map response is "[object Object]..."

• add "prompt" param to crawl example

• I had the mcp on cursor trying to search for local files multiple times. Can we improve the description to make sure that doesn't happen?

• search needs [ "web" ] as default sources (and explain the option in the function description), otherwise will throw empty response.

• extract doesnt have systemPrompt in v2

• firecrawl_deep_research throws error "Unknown tool: firecrawl_deep_research"

• generate_llmstxt doesnt exist on v2 anymore: "client.generateLLMsText is not a function"

• firecrawl_scrape only works for markdown format, is that the expected bahavior? If you try to run firecrawl_scrape with formats: ["summary"] it returns "No content available". Maybe add firecrawl_scrape_summary tool?

[nickscamara]

@rafaelsideguide fixed all of the above

[nickscamara]

Published a beta, I think we are close.

@tomkosm @rafaelsideguide please check the behavior for crawl and see if it is right.