How guides work
Users connect through an OAuth app you register with each provider. A guide says how far that app is from “anyone can connect”, and what stands in the way.
Difficulty
The path to public launch, not the coding. Depends on the scopes you request.
- Difficulty: Self-serve
- Create the app, get keys, go live. No provider review stands between you and your users.
- Difficulty: Capped
- Works today, but with a user, connection or rate cap until you apply for more.
- Difficulty: Reviewed
- Someone at the provider reviews your app before the public can connect.
- Difficulty: Gated
- Needs a paid security audit, business eligibility, a partnership or a paid API tier.
- Difficulty: Unconfirmed
- Research couldn’t establish the path. Plan as if it’s Reviewed until you confirm.
Time
- Your work
- Hands-on time. Agents can shrink it.
- Waiting
- Provider review time. You can’t shrink it much.
- Unknown
- Shown instead of a guess.
Who does each step
- Agent
- An agent can do this once you’re signed in.
- You
- Needs a person: sign-in, legal, payment or judgement.
- Provider
- Out of your hands. You’re waiting on the provider.
Evidence
Every claim is tagged. Numbers link to sources; low confidence is marked.
- Tested
- Observed during a setup attempt.
- Docs
- Stated in the provider’s official documentation.
- Community
- Reported by developers; not confirmed by the provider.
- Inferred
- Reasoned from related facts. Treat as a hypothesis.
Guide quality
- Tested
- The agent attempted setup. What happened is in the attempt log.
- Sourced
- Claims cite provider docs. Setup wasn’t attempted.
- Partial
- Research found little. Most claims are inferred.
For agents
/api/mcp- MCP server. Tools:
get_guide,research_app,research_status,list_guides,report_problem,report_setup. /apps/:slug.md- The guide as Markdown. Or send
Accept: text/markdown. /llms.txt- Index of every guide.
?format=prompt- On
/api/guides/:slug: steps an agent can act on, with stops for you.