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: Squad (formerly MissionControlHQ)iShort answer
In Squad, agents get 57 built-in research and scrape tools spanning web pages, search, people and companies, SEO, review sites, local businesses, and social. Zero API signups. Everything runs on credits (1,000/month included, $5 per 5,000-credit pack), and cost control is built in: per-mission budget caps, spend attribution by tool, agent, task, and mission, and optional auto-recharge with a monthly ceiling.
Key takeaways
| Principle | What it means in practice |
|---|---|
| No API-key project | 57 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-mission | 'Spend no more than 100 credits on this' is enforced |
| Spend is attributable | Usage breaks down by tool, agent, task, and mission |
Free access, capped spend: both at once.
The tools are built in
Web page, search, people and company, SEO, review, local, and social tools 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-mission caps keep any single job inside its allowance.
The spend is legible
By tool, agent, task, and mission, with a low-credit notification 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 tools cover
The coverage is the public web as an operator actually uses it:
- Social: social profiles and posts, including LinkedIn, where competitor and market signals surface first.
- Search: web search, for discovery and monitoring.
- Web pages: page scrapes from any URL, for everything else.
- People and companies: person and company lookups, for lead lists and account research.
- Local and reviews: local business listings, and reviews from review sites for reputation and complaint mining.
- SEO: keyword and ranking data, for content and competitor research.
The count sits at 57 tools; the per-tool price list lives in Settings > Billing, on the CREDIT COST PER USAGE card.
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 a new market, the tools for it already exist in every agent's hands.
The credit model
Everything runs on one balance: every workspace gets 1,000 included credits each calendar month (unused included credits expire at month end; they don't stack), purchased credits carry over from month to month while your subscription is active, 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 then hit the cache.
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, caps, and attribution
The fear that keeps operators from letting agents research freely is the runaway: a loop that scrapes 4,000 pages overnight. Two controls retire it:
- Per-mission budgets. Tell an agent "don't spend more than 100 credits on this research" and it records a budget for that mission; Squad refuses any research call that would take the mission over it. The agent sets the budget, so it can also change it, and there is no dashboard control for it.
- Attribution. Settings > Billing breaks credit spend down by tool, agent, task, and mission, alongside recent credit usage. When the balance drops below 1,000 credits (or your auto-recharge threshold), Squad sends a "Credits are running low" notification.
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's research calls are refused mid-task. Auto-recharge closes it: set a threshold ("when credits drop below 1,000"), how many packs to buy each time, and a monthly maximum so spending cannot run away. When the balance dips below the threshold, the top-up happens automatically; if the monthly maximum is reached, recharging stops for the rest of the calendar month. 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 social accounts, and two 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 Billing's usage by agent. The budget line in the lane's instructions ("cap any single mission 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? | Social + search tools | Sentiment lives in threads, not search snippets. |
| What do customers complain about? | Review-site tools | Clustered 1-2 star themes are positioning and roadmap input. |
| Unbounded research task | Mission budget in the instructions | 'No more than 100 credits' is enforced, 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 a mission budget
Do scheduled lanes depend on credits?
- If yes→auto-recharge with threshold + monthly cap
- If no: interactive research only→manual packs when the low-credit notification arrives
Frequently asked questions
The tools
Do AI agents need API keys to scrape the web? Not on Squad (formerly MissionControlHQ): 57 built-in research and scrape tools, on by default, cover web pages, search, people and companies, SEO, review sites, local businesses, and social. No API signups, 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: every workspace gets 1,000 included credits each calendar month (unused included credits expire at month end; they don't stack), purchased credits carry over from month to month while your subscription is active, 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 mission budget the agent records when you ask ('spend no more than 100 credits on this'), the auto-recharge monthly maximum as a hard ceiling, and credit usage broken down by tool, agent, task, and mission 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's research calls are refused mid-task for lack of credits.
Can I see which agent is spending the credits? Yes: Settings > Billing breaks usage down by tool, agent, task, and mission, alongside the balance and recent credit usage, and Squad sends a 'Credits are running low' notification when the balance drops. Spend is attributable, not a mystery number.
Sources
- Squad: homepage, get started, what uses credits
- Related on this site: Why Squad, Build prospect lists with AI agents, Automated competitor monitoring
Last updated: September 2026. Tool coverage and credit pricing checked against the Squad homepage and docs as of September 2026.