Docs
Generate your setup message, send it to your AI, and start.
Quickstart
Generate your setup message in Get started in the console, then send it to your AI (Claude, Claude Code, Codex, Cursor and more):
set up https://prolevelmax.com/SKILL.md
API key: sk_procli_…
Your AI starts fetching data as soon as it receives the setup message. New users get $2: $1 for verifying your email and $1 for your first data (once per person). No card required.
For apps that cannot run commands, such as ChatGPT, add the MCP connector. For your own code, use the REST API.
Direct collection
You can also collect data in the console without connecting an AI. The platform runs the collection for you.
- Open Data sources in the console and find a source by platform or keyword. Click its name to see the returned fields, uses and limits.
- Click Collect and enter the parameters (* marks required fields).
- Click Run and confirm the maximum charge. The actual charge never exceeds it.
Skill
For AI that can run commands or send HTTP requests, such as Claude, Claude Code, Codex and Cursor. Generate your setup message in Get started in the console and send it to your AI; the AI completes the setup. AI that can run commands save the key, so you only send it once.
set up https://prolevelmax.com/SKILL.md
API key: sk_procli_…
Example: set up https://prolevelmax.com/SKILL.md followed by API key: sk_procli_… on the next line. Keys are shown once. For ChatGPT, use the MCP connector below.
MCP (OAuth 2.1)
Remote MCP server with OAuth 2.1 and PKCE. An API key can also be used as a Bearer token.
https://prolevelmax.com/mcp
Confirm in Cursor or VS Code, then sign in to ProLevelMax and click Allow. No API key required.
Claude.ai / Claude Desktop: Customize → Connectors, search for ProLevelMax and click Connect. Or choose Add custom connector and paste the URL above.
ChatGPT (web): open Plugins, click + → Add custom MCP server, enter the URL above, choose OAuth, create it, then sign in to ProLevelMax and click Allow. In a chat, type @ and pick ProLevelMax.
Claude Code: add the server, then run /mcp inside Claude Code to sign in.
Codex: pass your API key through an environment variable.
Cursor, manual setup with an API key instead of sign-in: Settings → MCP → Add New MCP Server, with this config:
REST API
Authenticate with your API key as a Bearer token. Requests and responses are JSON. Machine-readable reference: openapi.json (OpenAPI 3.1).
| Endpoint | What it does |
|---|---|
POST /v1/discover | Find sources for a request in English or Chinese. Body: query, limit (1–50) |
POST /v1/inspect | Input fields, price and quote for one source (quote.maxCharge is the most the run can cost). Body: scraper, optional input and maxItems |
POST /v1/run | Run a source, billed to your balance. Body: scraper, input, maxItems (per-result sources return up to 100 unless set), timeout (seconds, max 300), async (true returns 202 and a runId at once) |
GET /v1/runs | Recent runs, newest first. Query: limit (1–100, default 20), offset |
GET /v1/runs/:id | Status, cost and results of a run. Query: offset, limit (1–1000) to read the results in pages |
GET /v1/runs/:id/download | Download the results. Query: format=csv, xlsx or json (nested fields become a.b columns) |
GET /v1/whoami | Verify the key and check the balance |
Pagination: per-call sources return one page per call. The result's page object contains the cursor for the next page (for example cursor, after or max_cursor); pass it as the paging field listed by inspect.
Schedules
Run a source automatically every day or every week at a set Beijing time, for example to monitor Xiaohongshu for brand mentions each morning. Each run is billed as a standard run and does not exceed its runCap; monthlyCap optionally limits spend per calendar month. Results appear in the run history. After 3 consecutive failed runs, or when the balance is insufficient, the schedule pauses and sends an email notification.
| Endpoint | What it does |
|---|---|
POST /v1/schedules | Create a schedule. Body: scraper, input, maxItems, frequency (daily or weekly), weekday (1–7 or mon–sun, weekly only), time (HH:MM, Beijing time), runCap (USD; default: the source's quote for this input), monthlyCap (optional), notifyEmail. With dryRun: true it returns a preview and saves nothing |
GET /v1/schedules | All schedules: next run, last run (runId and status), spend this month, enabled status |
GET /v1/schedules/:id | One schedule |
PATCH /v1/schedules/:id | Update the time, caps or email notification, or enable and disable it with enabled; re-enabling clears a pause |
DELETE /v1/schedules/:id | Delete a schedule; past runs are kept |
MCP tools: procli_schedule_create (call without confirmed to get a preview, then with confirmed: true after the user approves the cost), procli_schedule_list, procli_schedule_delete. Schedules can also be viewed, paused and deleted in the Dashboard under Schedules.
Tools
Tools available over MCP.
procli_discoverFind sources for a request, in English or Chineseprocli_inspectInput fields, price and quoteprocli_runRun a source (charges your balance); returns a preview, the total, the cost, the balance and download linksprocli_runs_getRun status and results, in pagesprocli_balanceCheck the current balanceprocli_schedule_createSchedule a daily or weekly run (previews the cost; saved only after the user confirms)procli_schedule_listList schedules with their next and last runsprocli_schedule_deleteDelete a scheduleBilling
- Two modes: per call (a fixed price per call, one page per call) and per result (billed by results returned, plus a run fee and a per-run minimum). Each platform's mode is listed on the pricing page.
- inspect returns a quote (the maximum the run can cost) before the run. The charge does not exceed the quote. The AI reports the price first and requests confirmation before any run that can cost more than $0.10.
- Per-result sources return up to 100 results unless maxItems is set. Run size is capped by the balance; a reduced run is noted in its result. Accounts cannot be overdrawn.
- Failed runs are generally not charged. If part of the work completed before the failure, a partial charge applies, not exceeding the run's quote.
- Each account can run up to 5 jobs concurrently (20 for Enterprise accounts) across all of its API keys and connections. Results are retained for 30 days.
- Refunds: all top-ups are final and non-refundable. To dispute a run's charge, open the run and select “Report a billing problem”; confirmed errors are credited back to your balance. See the refund policy.
Errors
Errors are JSON: {"code": 402, "type": "insufficient_balance", "message": "…", "requestId": "…"}. type is a stable identifier you can branch on.
| HTTP | type | Meaning | Resolution |
|---|---|---|---|
202 | Async run started | Poll GET /v1/runs/:id with the runId | |
400 | invalid_request | A required field is missing or malformed | Check the request body against the docs |
401 | unauthorized | Missing or invalid API key | Check the Authorization header |
402 | insufficient_balance | Insufficient balance | Top up in the dashboard |
402 | spend_limit | The API key or connected app reached its daily or monthly spending limit (UTC) | Raise the limit in the Dashboard under API keys or Settings → Connected apps; no top-up needed |
403 | email_unverified | The account email is not verified | Verify the email in the Dashboard under Get started |
404 | unknown_source · run_not_found · not_found | Unknown source or run ID, or the requested post or user does not exist | Use discover to find valid sources, or check the ID |
405 | method_not_allowed | Wrong HTTP method | See the Allow header |
409 | balance_reserved · run_in_progress | Balance held by runs still in progress, or the run has not finished | Retry when they finish, or ask for fewer results with maxItems; no top-up needed |
422 | invalid_input | The source rejected the input | Fix the input using inspect's fields; do not retry unchanged |
422 | no_data | The source returned no data for this input (an ID or link that is invalid, deleted or private, or the source could not fetch it at the moment) | Check the ID or link; if it is correct, retry once later. Generally not charged; any charge stays within the quote |
429 | too_many_runs · rate_limited | Too many runs at once (5 per account, 20 for Enterprise accounts), or the source is busy | Retry after a short wait |
502 | source_error | Data source error | Retry. Generally not charged; any charge stays within the quote |
503 | unavailable | Source temporarily unavailable | Retry later or use another source |
Response changes
Changes to the shape of a source's results are listed here. Where possible the previous shape stays available.
| Date | Sources | Change | Previous shape |
|---|---|---|---|
| 2026-11-20 | All sources | While a source is temporarily failing, a backup that returns the same data and costs no more takes the run: the result then has routedFrom (the source asked for) and routeNote (what happened); its fields can differ from the source asked for. Without such a backup, a failed run returns 503 with a message saying since when the source is failing and whether the call was charged, and alternative naming a source to use. | Add "backup": false to a REST request to turn this off |
| 2026-10-23 | Reddit sources' filter_posts; discover | Values shorter than 4 characters in filter_posts are dropped instead of failing the whole call, and a single string is accepted. In discover, bulk sources whose smallest run costs $0.10 or more (such as procli_reddit_a) rank after per-call sources unless the request asks for volume, and their why states the per-run minimum. Fields are unchanged. | None |
| 2026-10-22 | MCP tool procli_run | A run whose quote is above $0.10 starts only with confirmed: true in the arguments; without it nothing runs or is charged and the result has code: "needs_confirmation" and the quote. Runs up to $0.10 and the REST /v1/run are unchanged. | After the user approves the price, call again with the same arguments plus confirmed: true |
| 2026-10-22 | All sources (currently procli_reddit_c, procli_reddit_d) | During an upstream outage a failed run returns 503 with type unavailable and a new alternative field naming a source that returns the same data; inspect then also includes status: "unavailable" and alternative. Successful results are unchanged. | Optional: read alternative and switch to that source (check its price first) |
| 2026-10-21 | procli_bluesky_a | Data now comes straight from the platform's public APIs, which is faster and more reliable. The price is now $0.003 per call, with up to 1000 posts per run. Result fields and their order are unchanged. | No change needed |
| 2026-10-21 | procli_hackernews_a | Data now comes straight from the platform's public APIs, which is faster and more reliable, and the price is now $0.003 per call. Result fields and their order are unchanged. | No change needed |
| 2026-10-20 | procli_reddit_c · procli_reddit_l · procli_reddit_q · procli_reddit_r · procli_reddit_s | Reddit feeds return one row per post (id, title, text, author, createdAt, url, score, commentCount), without promoted posts or recommendation blocks. | Add "format": "raw" to the input |
What to do when a data source has problems
1. Am I charged when a run fails?
Usually not. Calls that error out, time out or find no data are not charged. If a run gets some results before it fails, you only pay for what came back, and never more than the quote shown before the run. A few sources that charge a start fee per run may still charge the small actual cost when nothing comes back; the run result will say so.
2. What if a source keeps failing?
When we spot an outage, we post a notice in the console, and another one once it's fixed. If you use ProLevelMax through an AI assistant, it will switch to a backup source as prompted. If you call sources yourself, use discover (or the source list on the website) to find a "recommended" source for the same data. If that still doesn't work, please tell us in the support chat.
3. Why is a source marked "not recommended", or missing from the website?
When a source has been failing a lot lately or has a temporary outage, the system marks it "not recommended" and points you to other sources for the same data. It shows again automatically once it's back to normal.
4. Why did I get "no data"?
Usually the link or ID is wrong, the post or account was deleted, or the account is private. These calls are generally not charged, so just try another link or keyword.
5. How do I contact you?
Leave a message in the support chat in the console, or email toph@prolevelmax.com. We'll get back to you as soon as we can.