Context.dev Review 2026: Pricing, Features & Real Limitations

Who should consider Context.dev?

The strongest use case is software with a recurring web-data requirement: an onboarding product building company profiles, a research assistant gathering evidence, or a knowledge system importing documentation. A shared provider can reduce integration work. It does not establish that every endpoint is the cheapest or most accurate option.

Context.dev is developer infrastructure. A dashboard can help with evaluation, but production adoption still needs software or an automation that consumes the output. Its API overview distinguishes page reads, discovery, research, parsing and enrichment.

What does Context.dev actually do?

Scrape converts a known URL into selected outputs such as Markdown, rendered HTML, screenshots or structured data. Map discovers URLs. Crawl collects linked pages, while batches handle larger asynchronous jobs. Search starts with a query; Answers performs research into a requested answer shape. Brand retrieves company and visual identity data.

That distinction matters more than the number of features. If you already know the URL, discovery is an extra step you may not need. If you need several supporting sources, interpreting one page is a different job from researching the web. Choose the operation around the evidence your application requires.

Monitors can watch page text, sitemap membership or relevant changes across a site's key pages. Exact detection and semantic detection have different meanings: a text change and a change important to your instructions are not equivalent. See monitor targets and detection modes.

Recent changes that affect a buying decision

The October 20, 2026 news-search announcement adds a useful research path for company events and current reporting. Context.dev describes an index exceeding 200 million articles, with topic or company queries and filters. This is the vendor's stated coverage, not a completeness guarantee or our own measurement.

The October changelog also records batches of up to 100,000 URLs on October 15, new social-source parsers and updates to credit-exhaustion errors. Read dated entries as release history and check the current operation guide before implementing a limit. Existing accounts can retain different allowances during a rollout.

Highlights, introduced September 28, selects relevant passages for a query. It can reduce the material sent to a model, but shorter output can omit an exception. The provider publishes its own benchmark and methodology; those results do not establish performance on your corpus. Read the Highlights announcement and evaluation notes.

Context.dev pricing: plans, quotas and overage

The public monthly plan view lists the following. Annual billing is a different commitment. Check your own account before changing an existing subscription. Current prices and allowances.

Listed overage per 1,000 credits is $2.20 on Developer, $1.80 on Pro, $1.40 on Growth and $1.00 on Scale. Free has no overage. Logo Link has its own allowance; it should not be counted as additional API credits.

How far do 1,000 credits go?

These are separate, hypothetical uses of the same balance, with no other consumption. Browser actions, additional formats and other operations can alter the calculation. Ultra is the documented default for Answers. Partial results, cached outputs and some missing targets can be charged; HTTP status alone is not a billing rule. Operation costs and charge conditions.

When can Pro cost less than Developer plus overage?

Our calculation below uses the published monthly prices, enabled overage rounded up to 1,000-credit blocks, and otherwise identical settings. It excludes discounts, taxes and special account terms.

The first case uses three overage blocks; the second uses 43. This illustrates why you should choose by expected operation mix, not by subscription price alone. Plan features and throughput may justify a different decision even when the usage arithmetic favors the lower tier.

CSS parser, JSON extraction or Answers: which should you use?

The custom parser returns named fields and lists from page elements. Missing items become null and unmatched lists become empty arrays. It is a useful choice when you know which elements contain the values; it does not infer the meaning of every page automatically.

The JSON extraction guide describes a model reading page Markdown, JSON-LD and meta tags. Its warning about required fields matters: a required, non-nullable field may receive a best-effort value. For an absent price or unknown company attribute, allow a missing value instead of forcing a plausible-looking answer.

Answers uses an example object, not JSON Schema. Confusing the two contracts can produce a faulty integration even though both return JSON. An application should validate types, missing facts and the evidence for important values.

Does Answers prove every extracted field?

No. According to the sources and deadlines guide, its sources list is not a field-by-field citation map. It includes research resources and cited URLs under documented rules. If your application needs an attributable price, keep the price and its supporting URL together, then verify the page states that amount and billing period.

For example, “contact sales” is not a zero-dollar plan. An article naming a company is not necessarily evidence of its current headcount. These are validation choices in your application, not guarantees supplied by a well-formed JSON response.

Does Context.dev always return fresh data?

No. The current Scrape guide defaults to a maximum cached-output age of three days. Each output is cached separately, so a combined response can contain captures from different visits. Requesting the needed outputs together with maxAgeMs: 0 asks for a fresh capture. Inspect cache metadata when freshness changes the meaning of a result. Scrape freshness and caching.

Brand is different: supported lookups default to a three-month cache window, and its current guide also documents maxAgeMs: 0 for a refresh. Do not apply a remembered Brand rule to every web-data operation. Brand options and partial profiles.

For a pricing observation, save collection time, source URL, currency and billing interval. For a slowly changing brand palette, a cached response may be acceptable. The right freshness policy follows the product's promise to its users, rather than a blanket demand that every call be fresh.

Can it interact with pages and scrape behind a login?

Yes, with boundaries. Browser actions can click, scroll or wait before capture on paid plans. A failed action can leave affected outputs unsuccessful inside HTTP 200. Inspect the requested output, not only the request status. Browser actions and failures.

Paid browser profiles can save and reuse authenticated sessions. Profiles belong to the organization; concurrent use of one profile can return PROFILE_BUSY. Sessions may expire, and MFA or CAPTCHA can require interaction. Support for a feature is not proof that a particular portal will work. Saved browser sessions and error handling.

Evaluate authenticated collection only for access you are authorized to use. Where the workload depends on a long sequence of interactions, test the full sequence before treating an extraction API as a substitute for your existing browser automation.

Does zero data retention cover the whole platform?

No. ZDR must be enabled for the organization and requested on a supported call. The guide excludes Batches, Monitors, Brand retrieval and feedback from that request mode. Supported successful responses confirm application with X-Context-ZDR: true; your application still controls its own logging and storage. ZDR scope and confirmation.

This is a buying constraint, not a checkbox. If your design depends on large batches and zero retained request content, establish a compatible architecture before choosing the platform. ZDR also changes available outputs, including OCR and hosted files. Review the applicable agreement and controls for your actual workload.

MCP, API and scale: what changes in production?

The hosted MCP endpoint is https://mcp.context.dev/mcp. The connection guide documents OAuth and client-specific routes. Tool calls consume the connected organization's credits. Verify a real tool result and the selected account; a model saying that it searched is not sufficient evidence of a working connection.

For scheduled ingestion, a REST or SDK workflow may give your application clearer control over inputs and retries. Batches are asynchronous: a completed job means pages were attempted, not that every page produced a usable output. Batch lifecycle and results.

Check effective limits in the account. The rate-limit guide distinguishes organization concurrency from older per-minute plans. Your number of API keys should not be treated as permission to multiply shared capacity.

If your wider problem is discovering and routing tools from several providers, our Monid review evaluates that separate layer. It does not replace checking Context.dev's own outputs, quotas or data handling.

The limitations that matter most

• Structured output is not verified truth. Check evidence and allow unknown values.
• A fetch can succeed while an output fails. Validate the part the application requested.
• Freshness and cost are operation-specific. A standard scrape is not a researched answer.
• The free plan cannot prove every paid workflow. Browser features need an appropriately scoped evaluation.
• One provider does not remove application engineering. Storage, retrieval, output validation and user decisions remain yours.

A focused evaluation before you commit

Select pages and companies that represent your real workload, including a few difficult cases. Define required outputs in advance. For each request, record the operation, source, collection time, fields present, errors and credits. Manually review a sample of both accepted and rejected records.

Then repeat after a source changes. A product that imports successfully once may still struggle with refreshes or conflicting information. Compare the finished workflow with the implementation you would otherwise maintain. This is our proposed evaluation method, not a report of tests we have performed.

When should you choose another approach?

For a broad provider shortlist, the existing French web scraping API comparison separates discovery, extraction and browser infrastructure; consult current vendor terms for prices. This review stays focused on Context.dev rather than repeating that comparison.

A specialist may fit better if one narrow collection task dominates. A self-managed crawler may suit a team that needs operational control and can maintain it. A manual process may remain best for low-volume research. For AI citation visibility, use our guide to creating citable content; buying a data API does not itself make your website a source in AI answers.

Our verdict: useful breadth, provided the workflow fits

Context.dev deserves evaluation when your application benefits from combining web content, structured extraction and company research. Its current feature set is broader than the “scraping plus logos” description found in some older reviews.

The reason to adopt it is a verified improvement in your own workflow: fewer components to maintain, satisfactory evidence and output quality, acceptable freshness, and a cost your product can support. Start with the job you need done, choose the correct operation and validate the result before scaling.

Context.dev review: frequently asked questions

Sources and review method

Primary sources checked October 21, 2026. Recommendations are our documentary assessment. Calculations use stated inputs; vendor announcements and benchmarks remain attributed to their publisher. Features, account entitlements and billing terms can change.

1. Context.dev API overview
2. Plans, allowances and overage pricing
3. Operation credits, partial results and charge rules
4. Custom CSS parser: fields, lists and limits
5. JSON extraction: schema and missing facts
6. Answers: research and structured responses
7. Answers json_format and response validation
8. Answers sources, partial output and deadlines
9. Scrape freshness and independent output caches
10. Brand retrieval: profile and freshness settings
11. Browser actions: paid access and failures
12. Browser profiles: saved sessions and authentication limits
13. Zero data retention: supported operations and exclusions
14. Hosted MCP server and authentication
15. Rate limits: concurrency and older per-minute plans
16. Batches: asynchronous collection and per-page outcomes
17. News search announcement — October 20, 2026
18. Changelog: dated changes to product behavior
19. Highlights: relevant passages and vendor benchmark methods
20. Monitors: exact and semantic change detection