A research agent is only as good as its access to the live web, and historically that access was a project: sign up for a search API, a social API, a scraping service, then wire keys into every agent. Built-in scrape tools collapse the project into a capability that is just there, and the credit model around them answers the other half of the problem: how to let agents research freely without fearing the bill.
0 credits
is what a failed scrape or a cache hit costs. Agents can retry errors and re-read yesterday's pull freely; only successful live scrapes spend budget.
Source: MissionControlHQiShort answer
In MissionControlHQ, agents get 34 built-in scrape tools spanning X, YouTube, LinkedIn, Reddit, Instagram, TikTok, Google search and news, HackerNews, ProductHunt, GitHub, Bluesky, full pages, screenshots, and structured extraction, with zero API signups. Everything runs on credits (1,000/month included, $5 per 5,000-credit pack), and cost control is built in: per-task budget caps, cost previews before committing, per-agent spend attribution, and optional auto-recharge with a monthly ceiling.
Key takeaways
| Principle | What it means in practice |
|---|---|
| No API-key project | 34 tools work out of the box; agents just call them |
| Credits, not invoices | 1,000/month included; $5 per 5,000-credit pack |
| Failures and cache are free | Retries and re-reads cost 0; only live pulls spend |
| Budgets are per-task | 'Spend no more than 100 credits on this' is enforceable |
| Spend is attributable | Usage breaks down by category and by agent |
Free access, capped spend: both at once.
The tools are built in
Social, search, video, code, and page-level scrapers with no keys to manage.
The cache works for you
Repeat reads cost nothing; freshness: fresh bypasses it when live data matters.
The budget is enforced
Per-task caps and cost previews keep any single job inside its allowance.
The spend is legible
By category, by agent, per task, with an amber dot before you run dry.
The API-key project agents shouldn't need
Give an agent a research task and it needs sources: search results, social threads, competitor pages, video transcripts. The traditional route to those sources is a stack of vendor relationships: one API for search, another for social data, a scraping service for pages, each with its own signup, key, quota, and bill. For a developer this is friction; for the non-technical operator it is a wall, and it is why most "research agents" in the wild can only read what a search snippet shows them.
Built-in tools remove the wall entirely. The agent calls scrape tools the way it calls any other capability; the platform carries the vendor relationships behind the scenes; you sign up for nothing and paste no keys. The research quality difference is immediate: an agent that can pull the full thread, the actual transcript, and the rendered page reasons about reality instead of about snippets.
What the 34 tools cover
The coverage is the public web as an operator actually uses it:
- Social: X/Twitter, Reddit, Instagram, TikTok, Bluesky, LinkedIn (profiles, companies, posts, search).
- Video: YouTube videos, comments, transcripts, and channels: transcripts being the quiet workhorse for research lanes.
- Search and news: Google search and Google News, for discovery and monitoring.
- Builder ecosystems: HackerNews, ProductHunt, GitHub, where competitor and market signals surface first.
- Any page: full web-page scrapes, page screenshots, and structured extraction from any URL, for everything else.
The practical effect is that research tasks stop being scoped by what is accessible and start being scoped by what is worth reading, which is the correct constraint. It also means new lanes need no procurement step: the day you decide the squad should watch ProductHunt launches, the tool for it already exists in every agent's hands.
The credit model
Everything runs on one balance: the plan includes 1,000 credits per month (resetting on the 1st), and additional 5,000-credit packs cost $5 from the Billing page. Two pricing choices do most of the work of making this comfortable:
- Failed scrapes cost 0. An error is never a charge, so agents retry freely.
- Cache hits cost 0. Reading data the platform already has is free, so repeated reads of the same source (a competitor page checked by three different lanes, say) spend once and reuse forever after.
The result is that credits track new information acquired, not activity performed, which is the number you actually want to pay for.
Freshness: when the cache should lose
The cache's economics are great right up until staleness matters. Every scrape tool therefore accepts a freshness parameter: the default (any) uses cached results when available, while fresh bypasses the cache and pulls live, for breaking-news threads, live-event monitoring, or a pricing page you have reason to believe changed this morning.
This is a judgment you can encode in the lane itself: the weekly competitor sweep runs on any (a day-old cache is fine and often free), while the launch-day watcher runs on fresh (staleness is the one failure that matters). The parameter turns "how current is this data" from an unknown into an explicit per-task choice.
Budgets, previews, and attribution
The fear that keeps operators from letting agents research freely is the runaway: a loop that scrapes 4,000 pages overnight. Three controls retire it:
- Per-task budgets. A budget tool lets you tell an agent "don't spend more than 100 credits on this task", and the cap is enforced, not advisory.
- Cost previews. A preview mode returns what a scrape would cost before committing, so agents (and their instructions) can check price tags first: "give me a cost preview" is a sentence that works.
- Attribution. Credits are tracked per task, and the credits panel breaks spend down by category and by agent alongside recent transactions. The header carries a live counter that quietly dots amber when the balance runs low.
Together these make research spend look like any other operational budget: allocated in advance, visible during, attributable after.
Auto-recharge for scheduled research
Scheduled research adds one failure mode budgets don't cover: the balance hits zero at 3am and the lane silently pauses mid-task waiting for credits. Auto-recharge closes it: set a threshold ("when credits drop below 1,000"), how many packs to buy each time, and an optional monthly cap so spending cannot run away. When the balance dips below the threshold, the top-up happens automatically; if the monthly cap is reached, recharging pauses until the 1st. The cap is the important half: auto-recharge without a ceiling is a blank check, and the design here deliberately makes you write the ceiling down.
A research lane in numbers
A concrete month for a competitor-watch lane: the Monday sweep pulls six competitor sites (structured extraction), their X accounts, and two subreddit searches. First run: roughly 60 credits of live pulls. Weeks two through four: most page reads hit cache at 0, fresh pulls only where content changed, about 25 credits per week. One launch-day watch runs fresh all day and spends 80. Month total: around 215 credits, comfortably inside the included monthly 1,000 with headroom for a second lane, attributed cleanly to the research agent in the credits panel. The budget line in the lane's instructions ("preview costs; cap any single task at 150") never fired, which is the point: caps exist for the abnormal day, and cost the normal ones nothing.
Tool selection cheat sheet
| Scenario | Best pick | Why |
|---|---|---|
| Competitor pricing page check | Structured extract, freshness: any | Cache is usually fine and free; extract beats raw HTML. |
| Breaking-news or live-event watch | freshness: fresh | Staleness is the one failure that matters here. |
| What are people saying about X? | X + Reddit + HackerNews tools | Sentiment lives in threads, not search snippets. |
| Long-form source (talk, tutorial, review) | YouTube transcript tool | Transcripts turn an hour of video into readable input. |
| Unbounded research task | Budget cap + cost preview in the instructions | 'No more than 100 credits' is enforceable, not advisory. |
| Scheduled lanes that must never pause | Auto-recharge with a monthly cap | No 3am credit-starvation; the cap keeps it honest. |
Does staleness matter for this read?
- If no: weekly sweep, background research→default freshness; cache hits are free
- If yes: live events, launch days→freshness: fresh, budgeted
Is the task's research bounded?
- If yes: known sources, known depth→run it; attribution handles the rest
- If no: open-ended discovery→cap it with the budget tool + preview costs
Do scheduled lanes depend on credits?
- If yes→auto-recharge with threshold + monthly cap
- If no: interactive research only→manual packs when the header dots amber
Frequently asked questions
The tools
Do AI agents need API keys to scrape the web? Not on MissionControlHQ: 34 built-in scrape tools cover X, YouTube, LinkedIn, Reddit, Instagram, TikTok, Google search and news, HackerNews, ProductHunt, GitHub, Bluesky, full pages, screenshots, and structured extraction, with no API signups and no keys to paste. Agents call them like any other tool.
What does the freshness parameter do? It controls cache behavior: the default uses cached results when available (at 0 credits), while freshness set to fresh bypasses the cache and pulls live data, for breaking news, live events, or anything where staleness matters more than credits.
The economics
How is agent scraping priced? On a simple credit balance: the plan includes 1,000 credits per month (resetting on the 1st), and additional 5,000-credit packs cost $5. Failed scrapes and cache hits cost 0 credits, so retries and repeated reads never burn budget.
How do I stop an agent from overspending on research? Three controls: a budget tool that caps a task ('spend no more than 100 credits on this'), a preview mode that returns the cost before committing, and per-task credit tracking so the bill maps to specific work. Runaway research is a configuration choice, not a risk you accept.
What is auto-recharge and should I turn it on? A billing setting that tops up credits automatically when the balance drops below your threshold, with a monthly cap so spending cannot run away. Turn it on if agents run scheduled research: it prevents the failure where a lane silently pauses mid-task waiting for credits.
Can I see which agent is spending the credits? Yes: the credits panel breaks usage down by category and by agent, alongside the balance and recent transactions, and a live counter in the header dots amber when you run low. Spend is attributable, not a mystery number.
Sources
- MissionControlHQ: homepage, early access
- Related on this site: Why MissionControlHQ, Mission Control for OpenClaw
Last updated: July 2026. Tool coverage and credit pricing verified against the live dashboard changelog as of July 2026.