AI Edge's seven-repo video makes a tempting pitch: add research, browsing, ads, diagrams and specialist agents to Claude in one sitting. The useful version is more selective. Each repository supplies a different kind of capability, and several can read accounts, operate a browser or change your global setup. Pick the one that solves a job you already have; test it with a limited scope before stacking another.
If you need to understand a market conversation, try Last30Days on a topic you know and check its cited posts. If you need to test your own site, try Playwright MCP on a staging page. If your process is unclear, use Archify to draw it before automating. Claude Ads, gstack, Agency Agents and Buzz are better when you already know the workflow, owner and data boundary they will serve.
Watch AI Edge Test the Seven Tools
Source note, 27 October 2026: AI Edge published the video on 16 September 2026. The reader-supplied companion Markdown guide says its commands came from repository READMEs on 17 September. I checked the seven official repositories again for this article; where instructions differ, the current README takes precedence. The examples below are the creator's demonstrations, not independent tests of the installed tools. I did not install the seven repos, connect an ad account or verify the video’s combined GitHub-star claim. The business prompt is an original editorial exercise.
The Seven Repos, Organized by the Work
| Repo | Job it adds | First constraint to check |
|---|---|---|
| Last30Days | Recent social and web research. | Which sources actually ran; optional keys change coverage. |
| Playwright MCP | Browser navigation and site QA. | Browser profile, logins and permission to submit actions. |
| Claude Ads | Evidence-led paid-media review. | Account scope, data quality and human approval for changes. |
| Archify | Architecture and process diagrams. | Whether the input describes the real process accurately. |
| gstack | Idea, plan, code and release reviews. | Installation prerequisites and global instructions. |
| Agency Agents | Reusable specialist role instructions. | Pick a few roles; files are not trained models or employees. |
| Buzz | Shared channels for people and agents. | Early-stage app, relay setup and data access. |
Free repository does not mean free operation. Claude usage, optional data providers, account connectors, hosting and staff review may still cost money. GitHub stars tell you a project is visible, not that its installer, outputs or permissions have been audited.
Research the Conversation; Then Check the Browser's Actions
1. Last30Days: a signal finder, not a market verdict
AI Edge asks Last30Days what users say about Claude, ChatGPT and Gemini, then turns to complaints about paid communities. The project README describes a sweep across recent community and web material. It supports a Claude Code plugin and other skill hosts. Its optional API keys affect source coverage; without them, do not write as though every named social network was searched. Open the cited posts, note their dates and separate vocal users from representative customers. A recurring complaint is a lead for interviews, not proof of demand.
2. Playwright MCP: useful browser reach, real browser authority
At 05:04, the creator asks the browser to compare recent videos from his channel and a competitor. Microsoft's Playwright MCP server gives a connected agent structured page access and browser actions. That is useful for your own mobile QA or a public-page comparison. It is not inherently read-only: a logged-in browser can also submit forms or change settings. Start with public pages, a separate test profile and explicit “do not submit” boundaries; verify the extracted numbers against the live page.
Audit Ads and Explain the Process Visually
3. Claude Ads: distinguish findings from account changes
The paid-media segment frames Claude Ads as an account analyst. Its current README describes evidence coverage, explicit unknown results and read-only adapters by default. Changes require a separately enabled capability, exact object IDs, a before/after diff and owner approval. That design is more useful than a magic health score: request a dated export or scoped access, review the conversion definition and ask for the three strongest findings with supporting rows. A score cannot establish that an ad change will raise revenue.
4. Archify: diagram what you can verify
Archify turns a description or codebase into an interactive diagram. AI Edge uses it around 11:59 to visualize a process. A good first use is a customer handoff or data flow you already know: name the systems, decision points and owner of each step, then have that owner correct the diagram. Attractive lines do not verify that a missing approval, exception or data transfer is absent. The official README documents the current skill installation and supported hosts.
More Roles Do Not Automatically Mean More Judgment
5. gstack: a review process with setup consequences
AI Edge uses Garry Tan's gstack to pressure-test a paid-community idea. The most transferable part is the sequence: question the idea, write a plan, review it from product and engineering angles, test the result. Its current install section lists a newer tested Bun version than the September guide and no longer asks you to paste a long hard-coded skill list into CLAUDE.md. Read the current install steps and inspect changes to global instructions before accepting them. For a small decision, simply borrow the questioning pattern rather than installing the full release workflow.
6. Agency Agents: specialist instructions, not trained staff
In the Agency Agents demo, AI Edge selects design, marketing and finance roles. The repository provides agent definition files that an existing assistant reads; its installer supports choosing only certain divisions or even a single role. The video calls them “trained” agents, but installing a Markdown persona does not train new model weights or grant professional accountability. A finance-role output still needs your data, an explicit question and a qualified human reviewer before it becomes a financial decision.
7. Buzz: a place to coordinate, not proof of autonomous work
Buzz is Block's open-source communication platform for people and agents. The creator demonstrates a channel and agent handoff at 22:40. The README describes both a relay you run and connections to an existing relay; a hosted community is not the only route. This is a young desktop product, and its Windows build is described as unsigned. Do not bypass a platform warning reflexively. Test with non-sensitive material, one agent identity and a review of what events are logged, who can see them and what the agent can access.
Use the Guide as a Map, Not a Bulk Installer
The supplied Markdown guide has a convenient seven-tool setup block, but it was assembled from September READMEs. Running it wholesale would register a browser server, install global skills and agents, alter global Claude instructions and download an app in one session. Current project pages are the authority for commands and prerequisites. A more useful first session is:
- Choose one job and one repo. Decide what a successful output looks like before installing anything.
- Open its official README and release page. Check the maintainer, license, version, platform requirements, dependencies and exact files or settings the installer changes. Record the version or commit you chose.
- Use the smallest scope available. A project-local skill or one Agency Agents division is easier to evaluate than a global install. Avoid live ad-account access and persistent browser logins for the first run.
- Run one harmless test. Examples: one public research question, a staging-page screenshot, a synthetic campaign export, or a made-up process diagram. Compare the result with the original source.
- Keep or remove it. Note the model/API cost, what permissions remained active and whether the tool improved the job enough to justify maintenance.
For example, the Playwright MCP README currently shows claude mcp add playwright npx @playwright/mcp@latest. That command registers a browser tool; it does not make its future actions safe. The Archify README shows npx skills add tt-a1i/archify -g, where -g means global installation. Read what you are consenting to before running either one. For Last30Days, Claude Ads, gstack, Agency Agents and Buzz, follow the current linked README rather than a copied one-shot prompt.
A Small Business Pilot to Copy
Of these seven, Claude Ads has a credible service application: a narrowly scoped, human-reviewed paid-media audit. This original prompt helps design a pilot; it does not connect an account, promise a return or replace an experienced media buyer.
Test one evidence-backed ad audit
One buyer, one account, no live changes.
Video Chapters
Get oriented
Read and act
| Time | Topic |
|---|---|
| 05:04 | Playwright MCP: browser control |
| 09:05 | Claude Ads: paid-media audits |
| 11:59 | Archify: diagrams from a brief |
Plan and collaborate
| Time | Topic |
|---|---|
| 14:41 | gstack: challenge and review a plan |
| 18:22 | Agency Agents: specialist role files |
| 22:40 | Buzz: people and agents in channels |
| 24:21 | Closing thoughts |
Sources and Limits
- Creator material: AI Edge's 16 September video, its supplied transcript and the reader-supplied setup guide dated 17 September. The guide's instructions were treated as dated source material, not executed.
- Research and browser: official Last30Days and Playwright MCP repositories.
- Marketing and visuals: official Claude Ads and Archify repositories.
- Planning and teams: official gstack, Agency Agents and Buzz repositories.
Repository counts, star totals, installation commands, supported platforms and hosted offerings can change. Recheck the current README and release notes before using a tool with valuable files or accounts.