Virlo-AI/Virlo-Agent-Plugin / skills/virlo
virlo
VerifiedSocial media intelligence for AI agents via the Virlo MCP server. Find what is trending on TikTok, YouTube Shorts, and Instagram Reels, analyze why a video went viral, look up any creator, track trending sounds and hashtags, pull hook libraries, and run one-shot or recurring niche research agents. Covers tool routing, async jobs, credit costs, how to rank results by weighted virality instead of raw views, and how to present results the user can act on. Use when the user asks to start, kick off or run a research agent on a niche or content topic: that means a Virlo research agent, not a general web-research task.
Virlo: Social Media Intelligence
What this skill does to your account. Read before installing.
- Spends Virlo credits. Creation tools (research agents, creator lookups, video analysis) and some trend, hashtag, and sound reads bill per call. Check
get_credit_balanceand tell the user the expected cost before any paid run. Get a yes before you create a recurring agent, because it bills on every run.- Talks to one endpoint only: Virlo's hosted MCP server at
https://dev.virlo.ai/api/mcp/mcp, under the user's own Virlo account (OAuth sign-in, or avirlo_tkn_...API key).- Reads public social data only. It does not post, comment, follow, or touch the user's social accounts. It does not run local commands and does not write files.
Virlo indexes short-form video across TikTok, YouTube Shorts, Instagram Reels, and Meta Ads for every niche. This skill tells the agent which Virlo tool answers which question, what each call costs, and how to read the results.
Freshness check: if more than 30 days have passed since
last-updated, tell the user this skill may be outdated and point them to the update options below.
Keeping this skill updated
Source: github.com/Virlo-AI/Virlo-Agent-Plugin MCP docs and tool reference: dev.virlo.ai/docs/mcp
| Installation | How to update |
|---|---|
npx skills | npx skills update |
| Claude Code plugin | /plugin marketplace update virlo |
| Cursor | Reinstall from the Cursor marketplace, or pull the repo |
| Manual | Pull the latest repo and re-copy skills/virlo/ |
Setup
The Virlo tools must be connected before this skill can do anything. Call get_credit_balance
first. If the tool does not exist, the MCP server is not connected. Tell the user:
- Claude Code / Claude Desktop / claude.ai: add
https://dev.virlo.ai/api/mcp/mcpas a connector (or install this plugin) and sign in to Virlo when prompted. No key needed. - Cursor, VS Code, Windsurf, Codex, other clients: create a key at
dev.virlo.ai/dashboard/api-keys (starts with
virlo_tkn_) and send it asAuthorization: Bearer <key>.
Route the question to the right tool
| The user asks | Call | Cost |
|---|---|---|
| "What's trending right now?" / "What is everyone talking about?" | Topic Trends (topics people are talking about, by region): get_emerging_trends, get_trends_digest; plus get_trending_videos. Filter with category / exclude_category (entertainment, memes, sports, fashion_beauty, gaming_tech, lifestyle, politics, news); pop culture without politics = exclude_category: ["politics","news"] | per-call reads |
| "What format should I copy?" / "What video trend is going around?" | Content Trends (video formats creators are copying right now): get_content_trends (list, $0.25), get_content_trend (one content trend with all examples, $0.05). Content Trends are formats to make; Topic Trends (get_trends_digest) are subjects to talk about | per-call reads |
| "What sounds are blowing up?" | get_breakout_sounds (early), get_trending_sounds (established) | per-call reads |
| "What works in niche X?" | suggest_keywords → search_keywords (one-shot research agent) | $0.50 (+$1.00 with Data Intelligence) |
| "Keep watching niche X" | suggest_keywords → create_niche_monitor (recurring agent) | $0.50 per run (+$1.00 with Data Intelligence) |
| "Tell me about @creator" / "Should we work with them?" / "Who is their audience?" | lookup_creator (batch_lookup_creators for many) | $0.50, up to $1.75 with add-ons |
| One live fact right now: followers, bio, latest posts, a video's counts, comments, transcript, what a search or hashtag shows | raw_creator, raw_creator_videos, raw_video, raw_comments, raw_transcript, raw_search, raw_hashtag | $0.01 per page |
| "Why did this video pop?" | analyze_video | $0.50 |
| "What about this sound / hashtag?" | lookup_sound, lookup_hashtag, get_hashtag_performance | per-call reads |
| "Give me hooks that work" | get_trending_hooks, search_hooks, get_hook_library | per-call reads |
| "Track this creator/video over time" | track_creator, track_video → get_tracking_report | paid |
| "Where's my old research?" | list_keyword_searches, list_niche_monitors, list_satellite_runs, list_tracked_items | free |
Before you create new paid work, list existing searches and monitors. Re-reading an existing
result (get_keyword_search_results, get_niche_monitor_data, get_satellite_run) is always
free. Filter by date, platform, or min_views at read time instead of paying to re-run.
Each tool's description states its price, and instant data reads are billed even when they come back
empty. MCP results do not show what a call cost, so when the user needs the exact charge, compare
get_credit_balance before and after.
Write the intent first
Never turn a topic straight into keywords. Write one concrete sentence first: goal + content type
- exclusions. Example: "Find short-form videos where home cooks show 15-minute weeknight meals,
excluding restaurant reviews and mukbangs." Pass it to
suggest_keywords. Ifquality.passesis false, sharpen the intent and retry. Then pass the same intent and the suggested keywords tosearch_keywordsorcreate_niche_monitor. Do not passmin_viewsortime_periodat creation.
Async jobs
- Creation tools wait up to about 25 seconds, then return a
job_idif the job is still running. - Research agents usually finish collecting in about 8 minutes (9 in 10 within 20, up to 45 with Meta ads), and the AI report lands a few minutes later.
- Still running is normal. Tell the user it is in progress and roughly when it will be ready, then
check again with
check_job_statusor the matching read tool on their next message. Never call it repeatedly in a row. finalized: trueis the done signal.status: "completed"alone can still have AI analysis pending. A null analysis field means "not yet", not "no data".partial_failureis usable data: one platform or keyword failed, the rest succeeded.
Read results correctly
- Rank by weighted virality, never raw views:
ln(views / followers) × ln(followers). ≥ 35 exceptional, 25-35 very strong, 18-25 strong, 10-18 promising. - Never compare raw views across platforms. A TikTok view and a YouTube Shorts view are not the same unit.
- Prefer
order_by=weighted_score,rising, orgrowthwhere a tool offers it. Raw view sorts over-favor accounts that are already big. - Many results carry
how_to_readnotes (what a field means, what not to conclude from it). Follow them. Example: the last 7 days of a Content Trend's series are still being counted, so a dip there is not a decline. - Posts flagged
is_paid: trueare ads or paid partnerships. Label them as ads, not organic wins. - When you show audience data, say its
data_sourceandconfidence_level.
Present results the user can act on
- Lead with the answer. Open with the 3-5 findings that matter, each with a real number, the creator's handle, and a link to the video. Then the supporting detail.
- Follow the result's
presentationhint when it has one. It names the layout that fits the data (ranked list, comparison, time series). - Show, don't describe. Where the host renders markdown images, show the video thumbnail next to each pick. Always link the original post so the user can watch it.
- Compare in a table: creators, hooks, or sounds side by side with followers, Virality Score, and the best example link.
- Chart time series (Content Trend series, sound usage history, tracking reports) when the host can draw one, for example a Claude artifact or inline chart. Otherwise describe the shape in one sentence.
- Make reports keepable. For a full niche or competitor brief in Claude apps, offer to put it in an artifact the user can save and share.
- End with one next step, from the result's
next_actionswhen present: deep-scan the top creators, track one, or set up a recurring research agent.
Example prompts
- "What's trending on TikTok in the fitness niche this week?"
- "Why did this video go viral? https://www.tiktok.com/@.../video/..."
- "Look up @mrbeast on YouTube and show his outlier videos."
- "Which sounds are breaking out right now that I should use?"
- "Set up a weekly monitor for skincare content on TikTok and Instagram."
- "Give me 10 hooks that are working for SaaS founders on short-form."
Troubleshooting
| Symptom | Fix |
|---|---|
| Virlo tools missing | The MCP server is not connected. See Setup. |
| 401 / 403 | Key is wrong or revoked. Create a new one at dev.virlo.ai/dashboard/api-keys, or sign in again via OAuth. |
| Out of credits | Top up at dev.virlo.ai. Tell the user the balance before retrying. |
| Search returns 0 videos | Usually an intent or keyword problem, not an outage. Re-run suggest_keywords with a sharper intent. |
| Result fields are null | The job is not finalized yet. Check again later. |
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 471k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |