BrowserAct vs Nexscope

BrowserAct 23

Hosted reading. No status page found (0) and no readable incident history (5). The docs say there are no rate limits, and give concurrency by plan (2 free, up to 20 for paid API users) and a 7-hour task cap (8). No 429 or retry guidance, and no idempotency keys on task creation (0). No SLA (0). No beta label on the API or MCP server (10). 0 + 5 + 8 + 0 + 0 + 10 = 23.

Nexscope 39

Hosted lines, graded on the remote MCP server and the REST API at api.nexscope.ai. No status page is linked from the home page, llms.txt, the API quickstart or the MCP map, and none was found (0). With no page there is no incident history to read (5). The pricing page publishes requests a second and concurrency per plan, the same for API and MCP, with 1 and 1 on the trial, 3 and 2 on Starter, 5 and 4 on Growth, 10 and 8 on Pro (15). The quickstart calls 429 an 'execution admission rate limit' and says to honour Retry-After on 429 and 503 when present. No idempotency key exists. The repository says 'Do not blindly retry POST requests' and to check whether a task was created before resubmitting, and the MCP tool pages say a reconnecting client should query existing async work rather than resubmit. SoldFetch earned 15 with idempotency keys and SandBase 11 with keys on some writes, so guidance without keys earns 9 (9 of 15). No SLA, and the terms say 'We cannot guarantee the Service will be available at all times' (0). No beta or preview label on the API or the MCP server (10). 0+5+15+9+0+10 = 39.

BrowserAct 76

An OpenAPI 3.0.3 document behind the API reference (25). llms.txt on both the site and the docs, with Markdown pages (10). Endpoint descriptions explain the run flow (find a bot, read its input_schema, then run) and when to use the lighter status endpoint (14). Inputs are typed, but a run's input is an object shaped by each bot's own schema (10). Examples per endpoint and named error codes such as 10879 and 10127; no full error table found (10). The path carries v3, but no changelog was found (7). 25 + 10 + 14 + 10 + 10 + 7 = 76.

Nexscope 64

A public OpenAPI 3.1 file at api.nexscope.ai/openapi.json, about 3.8 MB, linked from the quickstart. Visible request bodies are typed with required fields, patterns, enums and additionalProperties: false, but responses are open data objects and the components section, with the security scheme and the shared error response, wasn't reached. In the repository 166 of 252 request definitions are field descriptors rather than JSON Schema. The hosted tools/list wasn't read. SoldFetch took 17 with untyped responses and SandBase 22 with components unread (19 of 25). llms.txt on www.nexscope.ai and Markdown references for every API in the repository (10). MCP tool descriptions are one line ('Find Amazon products by keyword and marketplace. Results come back as structured JSON.'), with no when-not-to-use. The repository's references carry long trigger text ('trigger this skill whenever their need involves real-time Amazon search results'), not guidance on when to pick another tool (10 of 20). Tool pages list types, defaults, enums and length limits, but the Amazon Search page marks no argument required and omits the keyword its own examples pass, and many limits sit in prose (8 of 15). Every API has a request sample, a response fixture and cURL, JavaScript and Python clients, though most fixtures are labelled historical. Errors are an HTTP table, two named codes (11209, 12001) and a semantic error response whose schema wasn't read. MCP errors aren't documented (10 of 15). Path versioning under /v1, MCP v2 beside a legacy v1, and a dated OpenAPI version label, but no public changelog. SandBase earned 7 on the same pattern (7 of 15). 19+10+10+8+10+7 = 64.

BrowserAct 25

No x402, MPP or L402 (0). Plan prices and credit costs per bot build, run and workflow step are public, but there is no direct per-call price (15). A free plan with 200 credits; whether it needs a card isn't stated (10). A person signs up and creates the key in the dashboard (0). 0 + 15 + 10 + 0 = 25.

Nexscope 35

No x402, MPP or L402 on pricing, the quickstart, the MCP map, llms.txt or the visible OpenAPI paths. Credits come from a subscription or a Data top-up (0). Plans and credits per plan are public without a login ($29 for 30,000, $55 for 75,000, $129 for 150,000), and tool pages show credits per call, such as Amazon Search from 21 credits with a recent average of 30. Many tools on the MCP map list no cost and pricing is marked 'dynamic-credits'. Manifold earned 20 with a cost for every tool (15 of 20). A Free Trial of 1,000 credits for 3 days with 'No credit card required', API and MCP limits included. Creative API keys need a subscription (20). A person signs in through the browser to approve OAuth or create a key, and no keyless or programmatic route was found (0). 0+15+20+0 = 35.

BrowserAct 49

Graded on the API, since the MCP tool list isn't published. Task results come back whole, with no field selection found (10). Lists of bots, templates and tasks are paginated (15). Errors carry numeric codes with documented meanings for a few cases (12). No idempotency key on task creation; webhooks and a status endpoint avoid polling loops (5). No official SDK in two languages; the CLI is Python (7). 10 + 15 + 12 + 5 + 7 = 49.

Nexscope 44

Read on the MCP server, with the REST API where a line covers it. The MCP map lists 382 tools and says the category manifests 'narrow metadata for your integration; the MCP connection still advertises the full catalog'. That is more than 30 tools with no toolsets, dynamic loading or read-only subset found (5). Search tools page by page (about 20 results), with filters such as marketplace, category node, sort and device per tool, but no general pagination model or field selection, and output size depends on each tool (12 of 20). By default business errors answer HTTP 200 with an outer code. The X-Nexscope-Http-Status: semantic header maps them to HTTP statuses, the OpenAPI file names codes such as DATA_IMAGE_DOWNLOAD_FAILED and CAPABILITY_BACKEND_UNAVAILABLE in field text, and MCP errors aren't documented. SandBase earned 16 with typed bridge errors (11 of 20). Data calls only read. No idempotency keys and no readOnlyHint or destructiveHint documented. Async jobs return a taskId with a poll tool and a rule not to resubmit, and 1688 order, payment-link, receipt and cancel calls need an explicit JSON true confirmation (8 of 20). Few required parameters with defaults on most. No official SDK package, only standard-library clients in the repository for JavaScript and Python, which SoldFetch earned 8 for (8 of 15). 5+12+11+8+8 = 44.