The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
docs	Ready	Preview, Comment	Jun 15, 2026 2:35am

@greptile

@cursor review

PR Summary

Low Risk
Additive integration and documentation; user-supplied API keys call a third-party API with no changes to core auth or data paths.

Overview
Adds a new Context.dev integration (API-key auth, BYOK) so workflows can call scraping, extraction, and brand APIs from one block.

The context_dev workflow block exposes 22 operations via an operation dropdown—web scrape/crawl/map/search/screenshot, JSON-schema extract, product and design-system pulls, NAICS/SIC classification, brand lookup (domain, name, email, ticker, transaction), and prefetch helpers—with conditional sub-blocks and param mapping into the matching tools.

Implements 22 context_dev_* tools against api.context.dev, shared utils for auth/errors/credits, typed outputs (including file for screenshots from hosted URLs), and registers the block and tools in the sim registries.

Docs and catalog updates: integration MDX, meta.json nav entry, integrations.json entry, ContextDevIcon in docs/sim icon maps, and block meta with nine starter templates.

^{Reviewed by Cursor Bugbot for commit 3ff8e33. Configure here.}

Greptile Summary

Adds a full Context.dev integration block with 22 tools spanning web scraping, structured extraction, brand intelligence, industry classification, and utility prefetch endpoints. The implementation follows the existing Firecrawl pattern: AuthMode.ApiKey, BYOK API key forwarded via Bearer header, and shared output constants reused across related tools.

• 22 tools registered in tools/registry.ts and declared in blocks/blocks/context_dev.ts; a single block config with a 22-option operation dropdown routes all calls through one params dispatch switch, with separate maxPages subblocks for crawl (1–500) and extract (1–50) to avoid cross-contamination.
• Screenshot tool derives extension and MIME type from the returned URL with a known-extension allowlist, falling back to PNG — addressing the previous review finding.
• Shared utilities (utils.ts) centralise auth headers, error parsing, credit field extraction, and brand response transformation, keeping individual tool files lean; appendParam correctly forwards false booleans while skipping undefined/null/empty string.

Confidence Score: 5/5

New integration-only addition with no changes to existing tool or block logic; all 22 tools are isolated behind the context_dev namespace and cannot affect other integrations.

The integration is a well-contained additive change. Auth headers, error handling, credit extraction, and response transforms all follow the established Firecrawl pattern. The block's params dispatch is complete and verified (all 22 operations have matching switch cases, the API key uses user-only visibility, and the previously flagged includeFrames and screenshot MIME-type gaps have been resolved). The one naming ambiguity in classify_sic (input type vs output type) is a clarity issue, not a runtime defect.

apps/sim/tools/context_dev/classify_sic.ts — minor naming ambiguity between the type request param and the type response field worth tidying before this is consumed by LLM agents.

Important Files Changed

Sequence Diagram

sequenceDiagram
    participant User as User / LLM
    participant Block as ContextDevBlock
    participant Tool as Tool (one of 22)
    participant API as Context.dev API

    User->>Block: invoke with operation + params + apiKey
    Block->>Block: params dispatch (switch on operation)
    Note over Block: setBool / setString / setNumber coerce inputs
    Block->>Tool: resolved params object

    alt GET endpoint (scrape_markdown, get_brand, screenshot, map, …)
        Tool->>API: GET /v1/… ?params… (Bearer apiKey)
    else "POST endpoint (crawl, extract, search, prefetch_*, …)"
        Tool->>API: "POST /v1/… { body } (Bearer apiKey)"
    end

    API-->>Tool: "JSON { ...data, key_metadata }"
    Tool->>Tool: parseContextDevResponse → transformResponse
    Note over Tool: extractCreditMetadata(key_metadata)
    Tool-->>Block: "{ success, output: { ...data, creditsConsumed, creditsRemaining } }"
    Block-->>User: output fields

_{Reviews (3): Last reviewed commit: "fix(context-dev): wire includeFrames, sp..." | Re-trigger Greptile}

Greptile Summary

This PR adds a Context.dev integration with 10 operations spanning web scraping (markdown, HTML, screenshot, crawl, sitemap map), web search, structured extraction, NAICS/SIC classification, and brand data retrieval — modeled closely on the existing Firecrawl pattern (BYOK Bearer auth, appendParam-based URL building, shared credit metadata extraction).

• All 10 tools are correctly registered in both tools/registry.ts and blocks/registry.ts, with consistent param visibility (user-only for apiKey), shared utilities in utils.ts, and a unified block that dispatches to the right tool via an operation dropdown.
• The screenshot tool surfaces the captured image as a ToolFileData output so the executor's FileToolProcessor can download and store it as a UserFile, following the same path as other file-producing tools.
• Two capability gaps were found: includeFrames is declared in both scrape_markdown and scrape_html tool definitions but never forwarded in the block's params dispatcher or exposed in the UI; and the screenshot tool hardcodes mimeType: 'image/png' regardless of the format the API may return.

Confidence Score: 4/5

Safe to merge — the integration is self-contained, follows established patterns, and only minor capability gaps were found.

The implementation is consistent across all 10 tools, auth credentials are correctly scoped to user-only, and the file-output path for screenshots follows the established FileToolProcessor convention. Two small gaps exist: includeFrames is documented in both scraping tool definitions but never wired through the block dispatcher or UI, and the screenshot tool unconditionally labels files as image/png without inspecting the actual response content type. Neither affects correctness of the other nine operations.

apps/sim/blocks/blocks/context_dev.ts (params dispatcher completeness) and apps/sim/tools/context_dev/screenshot.ts (MIME type derivation).

Important Files Changed

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    User([User / LLM]) --> Block["ContextDevBlock\n(operation dropdown)"]
    Block --> Dispatch["params dispatcher\n(switch on operation)"]

    Dispatch --> SM["context_dev_scrape_markdown\nGET /web/scrape/markdown"]
    Dispatch --> SH["context_dev_scrape_html\nGET /web/scrape/html"]
    Dispatch --> SS["context_dev_screenshot\nGET /web/screenshot"]
    Dispatch --> CW["context_dev_crawl\nPOST /web/crawl"]
    Dispatch --> MP["context_dev_map\nGET /web/scrape/sitemap"]
    Dispatch --> SR["context_dev_search\nPOST /web/search"]
    Dispatch --> EX["context_dev_extract\nPOST /web/extract"]
    Dispatch --> CN["context_dev_classify_naics\nGET /web/naics"]
    Dispatch --> CS["context_dev_classify_sic\nGET /web/sic"]
    Dispatch --> GB["context_dev_get_brand\nGET /brand/retrieve"]

    SS --> FTP["FileToolProcessor\n(download → UserFile)"]
    FTP --> File[(Stored File)]

    SM & SH & CW & MP & SR & EX & CN & CS & GB --> Credits["creditsConsumed / creditsRemaining\n(key_metadata)"]

_{Reviews (2): Last reviewed commit: "feat(context-dev): add Context.dev web +..." | Re-trigger Greptile}

@greptile

@cursor review