Browse the full directory of Claude Code plugins — commands, agents, skills, MCP servers, and more.
Browse plugins →Connect Claude Code to databases, docs, and memory stores via MCP servers. 8 plugins from the directory's 7,552 MCP components show how.
MCP (Model Context Protocol) servers let Claude Code plugins expose tools that Claude can call during a coding session. Instead of relying solely on built-in capabilities, an MCP server can connect Claude to an external database, a documentation API, a vector store, or any other service that speaks the protocol.
The ClaudePluginHub directory currently indexes 7,552 MCP server components across 31,811 active plugins. That makes MCP the fourth-largest component type after skills (174,991), commands (59,131), and agents (47,060) — out of 296,432 total components. It's a relatively focused category: most plugins don't need an MCP server. But the ones that do tend to solve problems other component types can't address alone — persistent memory, live documentation lookup, and cross-agent coordination.
Some plugins ship MCP as their only component type. claude-flow and mempalace are both MCP-only: they exist to run a server process that Claude calls into. Others bundle MCP alongside commands, skills, hooks, and agents. The distinction matters when you're evaluating what to install.
An MCP-only plugin is lightweight — it does one thing and exposes it through a server. A multi-component plugin like ecc (commands, agents, skills, hooks, and MCP) uses its MCP server as one piece of a larger workflow. When you install ecc, the MCP server works alongside 300+ other components for coding workflows, TDD enforcement, security audits, and deployments.
This multi-component pattern is common among higher-install MCP plugins. Of the 8 plugins covered here, 6 bundle MCP with at least one other component type.
These 8 plugins represent three patterns: memory persistence, documentation retrieval, and agent orchestration.
Three plugins use MCP servers to give Claude persistent memory across sessions:
claude-mem (37 installs/7d, 76,498 stars) persists Claude Code context across sessions by capturing observations from Read, Edit, and Bash actions, storing them in a local SQLite database, and injecting relevant memories into new sessions. Components: skills, hooks, MCP. The hooks auto-capture context as you work; the MCP server makes that context queryable. It also includes skills for codebase mapping, structural search, PR monitoring, implementation planning, and release automation.
mem0 (56,038 stars) adds persistent memory via the Mem0 platform. It retrieves relevant past decisions, strategies, and session states when starting new tasks, and supports semantic search across long-term memories using Python and TypeScript SDKs, hooks, and MCP tools. Components: skills, hooks, MCP. Where claude-mem uses local SQLite, mem0 connects to an external service for memory storage and retrieval.
mempalace (52,460 stars) mines projects and conversations into a searchable vector store it calls a "palace." It runs a local MCP server that Claude queries via RAG tools, with auto-save functionality maintaining fresh indexes and guided setup for quick integration. Components: MCP only. This is one of two MCP-only plugins in this list, keeping the install footprint small if you only need the memory layer.
All three solve the same core problem — Claude Code sessions are stateless by default — but they differ in storage backend (SQLite vs. cloud platform vs. vector store) and in how automatically they capture context.
context7-plugin (55,548 stars) fetches up-to-date, version-specific documentation, API references, and code examples for libraries like React, Next.js, Vue, Prisma, and Supabase directly into Claude's context. Query via /context7:docs <library> [query] or by ID for precise lookups on setup, usage, and APIs. Components: commands, agents, skills, MCP. The MCP server handles the actual doc fetching; commands and skills provide the user-facing interface.
prompts.chat (162,464 stars) searches, retrieves, and manages thousands of AI prompts and Claude skills from the prompts.chat library. You can install skills to extend capabilities, fill prompt variables, save custom prompts with metadata, and enhance them using AI. Components: commands, agents, skills, MCP. The MCP server connects to the prompts.chat catalog, giving Claude access to a large prompt library without leaving the terminal.
claude-flow (53,330 stars) orchestrates a swarm of AI agents to execute complex, multi-step tasks with local and cloud-based coordination, leveraging 150+ commands and specialized agents for enterprise workflow automation. Components: MCP only. This is the other MCP-only plugin in the list — the server itself is the product, coordinating multiple agent processes through a single protocol endpoint.
ecc (255 installs/7d, 187,411 stars) is a comprehensive plugin with 300+ agents, skills, commands, and hooks for autonomous multi-agent coding workflows, TDD enforcement, security audits, and production code generation across JS/TS/Python/Rust/mobile stacks. Components: commands, agents, skills, hooks, MCP. The MCP server here is one piece of a much larger toolkit — useful if you want a single install that covers many workflows.
everything-claude-code (67 installs/7d, 2 stars) covers the same five component types — commands, agents, skills, hooks, and MCP. With lower star count and fewer installs than ecc, this is a newer entry worth evaluating against ecc if you want an alternative take on the multi-component approach.
Across these 8 plugins, several patterns stand out — both for developers building their own MCP servers and for those choosing which to install:
Hook + MCP pairing. claude-mem and mem0 combine hooks (to capture events automatically) with MCP servers (to make captured data queryable). If you're building a context-persistence plugin, this is the established pattern: hooks write, MCP reads.
MCP-only for single-purpose tools. claude-flow and mempalace ship MCP as their sole component. This keeps the plugin focused and avoids namespace collisions with other installed plugins. If your tool does one thing — serves documentation, proxies an API, coordinates agents — an MCP-only approach keeps the install clean.
MCP alongside commands for discoverability. context7-plugin and prompts.chat pair their MCP server with slash commands. Users can explicitly invoke /context7:docs rather than waiting for Claude to decide when to call the server. This dual-access pattern makes the tool both discoverable and automatic.
Multi-component bundles for workflows. ecc wraps MCP into a broader workflow toolkit spanning 5 component types. This makes sense when the MCP server isn't the main feature — it's infrastructure supporting a larger set of skills and agents. The tradeoff is a heavier install and more potential for component overlap with other plugins.
The directory's MCP servers page lists all 7,552 indexed MCP components, filterable by category and star count. For broader plugin discovery, the plugins directory covers all 31,811 active plugins across all 296,432 components.
If you're evaluating MCP plugins for a specific use case — memory, docs, orchestration — compare the component types each plugin includes. An MCP-only plugin installs lighter, while a multi-component plugin may give you a more integrated workflow at the cost of a larger footprint. The storage backend matters too: local SQLite (claude-mem) keeps data on your machine, while cloud-backed options (mem0) persist across devices but add an external dependency.
Skills make up 59% of all Claude Code plugin components. Learn how they work and explore 8 skill-based plugins — from TDD workflows to token-saving strategies.
How Claude Code hooks intercept lifecycle events, and which plugins use them for TDD enforcement, memory persistence, and token optimization.
58,466 slash commands span the Claude Code plugin directory. See how eight plugins use commands for workflow orchestration, code review, session HUDs, and prompt management.
Transforms Claude Code into a full-featured development platform with 270+ language-specific skills, 90+ CLI commands, and 67 specialized agents covering code generation, review, testing, debugging, deployment, security, and automation across 40+ technology stacks.
Complete collection of battle-tested Claude Code configs from an Anthropic hackathon winner - agents, skills, hooks, rules, and legacy command shims evolved over 10+ months of intensive daily use
Persist and retrieve conversational context across Claude Code sessions using a local filesystem-based memory store, enabling agents to recall past decisions, work patterns, and bugfixes without re-reading source files.
Search, retrieve, improve, and manage thousands of AI prompts and Claude skills from prompts.chat directly in your coding assistant. Install skills to extend capabilities, fill prompt variables, save custom prompts with metadata, and enhance them using AI.
Fetch up-to-date, version-specific documentation, API references, and code examples for libraries like React, Next.js, Vue, Prisma, and Supabase directly into your LLM context using Context7 skills, commands, agents, and MCP server. Query via /context7:docs <library> [query] or IDs for precise lookups on setup, usage, and APIs.
Orchestrate swarms of AI agents for complex multi-step tasks using SPARC methodology, swarm coordination, and GitHub automation, with WASM-accelerated local execution and a cloud-based orchestration platform providing 70+ tools.
Mine project files and conversation history into a local, searchable semantic memory palace, then query it during development to recall past work, decisions, and context without relying on model memory.