Agenta Plugin

Modular MCP plugin collection for Claude Code. Pick a profile, get a curated set of MCP servers and skills optimized for your workflow.

---

Quick Start

# Option 1: Interactive installer (defaults to "full" profile)
bash install.sh /path/to/your/project

# Option 2: Copy a template directly
cp templates/core.mcp.json /path/to/your/project/.mcp.json

# Option 3: Specify profile as argument
bash install.sh /path/to/your/project research

Note: If you cloned this repo, run git submodule update --init --recursive to fetch external skill repos (Solana Foundation, SendAI, Trail of Bits, Cloudflare, Superpowers).

---

Profiles

Profile	Use Case	MCP Servers	Key Tools
core	Any project	3	Sandbox, memory graph, context mode
knowledge-base	RAG, document management	12	Vector search, scraping, memsearch, Supabase, Postgres
research	Analysis, reports	11	Multi-engine search, scraping, academic papers, memsearch, Notion
web-dev	Full-stack development	12	Browser automation, scraping, memsearch, Supabase, Postgres
blockchain	Solana development	11	Helius, Jupiter docs, memsearch, context7, Twitter
full	Everything	21	All MCP servers combined

See docs/profiles.md for detailed profile descriptions and docs/guide.md for end-to-end user flows.

---

MCP Servers (21)

Server	Purpose	Profiles
e2b	Cloud sandbox (TS/Python)	all
memorygraph	Knowledge graph	all
context-mode	Context window management (98% token reduction)	all
claude-context	Vector search + code semantics	kb, blockchain
qdrant	Vector database	kb
sqlite	Local database	kb
supabase	Supabase (DB, auth, storage)	kb, web-dev
postgres	Direct PostgreSQL access	kb, web-dev
mcpdoc	llms.txt documentation	kb, research, web-dev, blockchain
crawl4ai	Web scraping	kb, research, web-dev
omnisearch	Tavily/Exa/Perplexity	research, web-dev, blockchain
context7	Live library docs	research, web-dev, blockchain
notion	Notion integration	research
consensus	Academic paper search (200M+ papers)	research
puppeteer	Headless browser	research, web-dev
daytona	Container sandbox	web-dev, blockchain
helius	Solana blockchain	blockchain
twitter	Twitter/X automation	blockchain
memsearch	Persistent memory with hybrid search	kb, research, web-dev, blockchain
skill-seekers	Source-to-knowledge ingestion	kb
clickup	Task management	full

See docs/mcp-catalog.md for setup details.

---

Skills (8)

Internal Skills

Skill	Description	Profiles
autonomous-agent	Task decomposition, memory, self-correction	all
knowledge-base	RAG patterns, ingestion workflows	kb
research	Multi-source search, synthesis	research

External Skills (Submodules)

Skill	Source	Description	Profiles
solana-foundation	Solana Foundation	@solana/kit, Anchor, Pinocchio, LiteSVM, Codama	blockchain
solana-ecosystem	SendAI	Jupiter, Drift, Raydium, Orca, 30+ DeFi protocols	blockchain
security	Trail of Bits	Static analysis, Semgrep, supply chain, secure contracts	blockchain
cloudflare	Cloudflare	Workers, Durable Objects, Wrangler, Agents SDK	web-dev
superpowers	obra/superpowers	TDD, debugging, code review, planning, git worktrees	web-dev

---

Project Structure

agenta-plugin/
├── profiles/          # Profile manifests (which MCPs + skills)
├── mcps/              # Individual MCP server configs
├── skills/            # Skill files and submodules
│   ├── *.md           # Internal skills
│   ├── SKILL.md       # Merged skill index
│   ├── solana-foundation/  # Submodule: Solana Foundation
│   ├── sendai/             # Submodule: SendAI
│   ├── trailofbits/        # Submodule: Trail of Bits
│   ├── cloudflare/         # Submodule: Cloudflare
│   └── superpowers/        # Submodule: Superpowers
├── templates/         # Ready-to-copy .mcp.json files
├── docs/              # Extended documentation
├── install.sh         # Profile installer script
├── CLAUDE.md          # Agent instructions template
├── plugin.json        # Plugin metadata
└── marketplace.json   # Plugin registry entry

---

Environment Variables

Each profile requires different API keys. See docs/env-vars.md for the full reference.

Minimum for each profile:

Profile	Required Keys
core	E2B_API_KEY
knowledge-base	E2B_API_KEY, OPENAI_API_KEY (+ optional: SUPABASE_URL, POSTGRES_URL)
research	E2B_API_KEY, TAVILY_API_KEY
web-dev	E2B_API_KEY (+ optional: SUPABASE_URL, POSTGRES_URL)
blockchain	E2B_API_KEY, HELIUS_API_KEY, OPENAI_API_KEY

---

How It Works

1. Choose a profile that matches your project type
2. Run the installer or copy a template .mcp.json
3. Set environment variables for the MCP servers you want to use
4. Start Claude Code — MCP servers and skills are automatically available

The installer copies:

• .mcp.json — MCP server configuration for Claude Code
• skills/ — Relevant skill files and submodule directories for your profile
• skills/SKILL.md — Generated index of all installed skills
• CLAUDE.md — Agent instructions (if not already present)

---

Submodules

External skills are tracked as git submodules. To update to latest upstream:

git submodule update --remote

If submodules are not initialized, the installer will attempt to init them automatically. If that fails (e.g., no git access), it will warn but continue — internal skills will still work.

---

Customization

Adding MCP sources to mcpdoc

The mcpdoc server supports custom llms.txt URLs. Edit your .mcp.json to add --urls arguments:

"mcpdoc": {
  "command": "uvx",
  "args": [
    "--from", "mcpdoc", "mcpdoc",
    "--urls", "MyDocs:https://example.com/llms.txt"
  ]
}

Creating a custom profile

1. Copy an existing profiles/*.json
2. Modify the mcps and skills arrays
3. Save as profiles/my-profile.json
4. Create a matching templates/my-profile.mcp.json

---

License

MIT License — see LICENSE

---

Links

• Agenta Labs
• Claude Code
• MCP Protocol