Categories
Shelves, not tags: every server sits in exactly one place. Assignment is automatic at sync time and corrected by hand as entries get curated.
Finance & PaymentsServers that touch money: payment APIs, market data, on-chain queries. Read the credential requirements twice — most of these want production API keys, and an agent with write access to a payment provider deserves real thought.Communication & MessagingSlack, email, chat and everything else that talks to humans. The good ones support read-only scopes; prefer those unless your agent genuinely needs to send on your behalf.Databases & Vector StoresStructured access to data stores — from Postgres to vector DBs. Favor entries with explicit read-only modes if your agent doesn't need writes, and check whether schema introspection is a tool or a config step.Search & RetrievalWeb search, site search, and retrieval backends for grounding agents in current information. Result quality varies more than the READMEs admit — the verified-recency stamp matters here more than stars.Browser & Web AutomationHeadless browsers, scrapers and page-interaction servers. The capability gap between 'fetches HTML' and 'drives a real browser' is huge — the capability matrix on each entry tells you which one you're getting.Media & ContentImage, audio and video: generation, transcription, editing, and platform APIs like YouTube or Spotify. Most generation servers proxy a paid API — check whose key you're burning before wiring one into a loop.Files & DocumentsFilesystems, object storage, and document formats — PDF, spreadsheets, Markdown, cloud drives. The workhorse category: unglamorous, heavily used, and where staleness hurts most because format edge cases rot quietly.Cloud & InfrastructureAWS to Cloudflare, Kubernetes to Terraform. These servers hold the keys to production — scope their credentials like you would a junior engineer's, because that's roughly what an agent with kubectl is.Productivity & Project ManagementNotion, Jira, Linear, calendars, tasks. Integration depth varies wildly — some expose the whole API surface, some expose three tools. The capability matrix saves you the disappointment.Developer ToolsGit forges, CI, error trackers, code intelligence. The category where MCP earns its keep fastest — an agent that can read your issues, your CI logs and your stack traces stops asking you to paste things.AI & Model ToolingModel routers, embedding services, agent frameworks, LLM utilities. Meta-tooling for the stack you're already in — and the category with the highest churn, so watch the stamps.Everything ElseEntries that don't fit a shelf yet — niche APIs, hardware bridges, experiments. Categories are reassigned as the index is curated, so this shelf shrinks over time.