The MCP Ecosystem Just Added 1,062 Servers in a Single Week
That's not a typo. The MCP catalog hit 23,812 total servers this week — and 1,062 of them are brand new. The pace isn't slowing down. If anything, the diversity of what's being built is getting more interesting.
This week's batch spans chemistry tools, zero-knowledge secret vaults, writers' studio integrations, and a memory layer that costs literally nothing per token. Let's get into it.
1. The $0/Token Memory Layer
io.github.YonasValentin/mcp-memory-graph is doing something genuinely clever: local-first persistent memory via a hybrid search + knowledge graph, with zero token cost. It scored a 91 — the highest of any new server this week — and works with Claude Code or any MCP-compatible client.
The "local-first, $0/token" framing is a direct shot at hosted memory solutions that meter every query. If you're running Claude Code heavily, this is worth an immediate look.
2. Chemistry Gets an Offline MCP Server
io.github.jurimaxam-dotcom/chemdraw-mcp converts molecule names or SMILES strings into 2D structures, reactions, mechanisms, and spectra — entirely offline via RDKit. Three stars already, 87 score total.
This is a niche pick but a significant one. Researchers who can't send proprietary compound data to a cloud API now have a viable path to AI-assisted chemistry workflows. Offline-first scientific tooling is an underserved corner of the MCP ecosystem.
3. The Vault That Keeps Secrets From the Model
io.github.wundervault/wundervault-mcp describes itself as a zero-knowledge secret vault for AI agents: secrets get injected into commands but are never exposed to the model itself. 86 score, zero stars yet — but the concept is sharp.
An AI agent that can act on your secrets without ever seeing them is a fundamentally different security posture than what most teams are running today.
This addresses a real problem: you want your agent to run authenticated commands, but you don't want API keys sitting in the model's context window. The architecture matters here, and the team deserves scrutiny — but the premise is sound.
4. Scrivener Finally Gets Serious MCP Support
Scrivener MCP ships 60+ tools for document management, writing analysis, content enhancement, semantic search, and character/plot tracking inside Scrivener. It works with Claude, ChatGPT, and other assistants.
Writers using Scrivener for novels, screenplays, or long-form research have been oddly underserved by the MCP ecosystem until now. Sixty-plus tools is not a shallow integration — that's a real attempt at making Scrivener AI-native.
5. Test Automation Across Every Stack
Mk Qa Master drives pytest, Jest, Cypress, Go, and Maestro from a single MCP server, covering web and mobile (iOS, Android, BlueStacks). At 36 stars and a 64 score, it's already finding an audience.
The breadth is the story here. Most testing MCP servers pick a lane. This one doesn't.
Web Scraping Is Having a Moment
Three of the top ten trending servers are web scraping tools — and the numbers are striking.
Firecrawl added 381 stars this week alone, sitting at 133,968 total. Scrapling added 276 stars on top of its 64,514 base. Puppeteer Core picked up 194 stars heading toward 95,022.
Something is clearly driving demand for clean web data pipelines. My read: as agentic workflows mature, the bottleneck shifts from "can the model reason?" to "can the model get good data?" Scraping infrastructure is becoming foundational agent plumbing.
The Claude Flow Duopoly
Both Claude Flow and Claude Flow MCP Server landed in the trending list — adding 145 and 136 stars respectively. Together they represent nearly 120,000 combined stars and a serious bet on multi-agent swarm orchestration.
The top-scored server in the entire trending list, at 94, is Claude Flow. Enterprise agent orchestration with 60+ specialized agents isn't a weekend project — and the market seems to be noticing.
When two variants of the same orchestration framework both trend in the same week, the underlying idea has escaped the early-adopter ghetto.
ByteDance Is Building Quietly
Both UI TARS Desktop (+140 stars) and io.github.bytedance/mcp-server-search (+137 stars) trended this week under the ByteDance umbrella. Combined star base: over 73,000.
A major tech company shipping open-source MCP tooling at this scale is worth watching. The search server is functional and straightforward; UI TARS is more ambitious — a multimodal agent stack connecting AI models to desktop automation. Neither is a toy.
What This Week Actually Tells Us
The 1,062 new servers aren't noise — they're signal. The categories breaking through this week (offline scientific tools, zero-knowledge secrets management, memory graphs, full-stack test automation) suggest builders are solving real production problems, not just prototyping.
The scraping trend reinforces something important: data infrastructure is becoming the competitive moat for AI agents, not the models themselves. The teams investing in clean, reliable data pipelines now are building advantages that won't be easy to replicate.
23,812 servers and counting. The ecosystem isn't maturing — it's accelerating.
Servers mentioned
MCP Security Weekly
Weekly CVE alerts, new server roundups, and MCP ecosystem insights. Free.
Keep reading
This article was written by AI, powered by Claude and real-time MCPpedia data. All facts and figures are sourced from our database — but AI can make mistakes. If something looks off, let us know.