AI Agent Tools List 2026: 30+ Tools Categorized for Developers and Builders
Building an agent is not hard for lack of tools; it is hard to know which layer to fill. This list covers 30+ tools across frameworks, memory, retrieval, browser automation, MCP, orchestration, and observability, plus a minimal stack.
💡 What You Will Learn
Building an agent is not hard for lack of tools; it is hard to know which layer to fill. This list covers 30+ tools across frameworks, memory, retrieval, browser automation, MCP, orchestration, and ob
📜 Table of Contents
AI Agent Tools List 2026: 30+ Tools by Layer
The agent stack has settled into layers - pick one or two tools per layer instead of guessing. Stars come from our tracker snapshot (late July 2026).
1. Frameworks
| Tool | Stars | Focus |
|---|---|---|
| LangChain / LangGraph | ~143k / ~39k | ecosystem and graph orchestration |
| AutoGen | ~60k | multi-agent conversation |
| CrewAI | ~56k | role-based crews |
| LlamaIndex | ~51k | retrieval-first |
| Semantic Kernel | ~28k | cross-language |
| Haystack | ~26k | production pipelines |
2. Memory
mem0 (~62k) compresses conversations into retrievable entries; Zep adds graph memory with entity extraction; Letta (formerly MemGPT) lets models edit their own memory. Choose by whether you need facts or relationships.
3. Retrieval
Milvus (~45k), Qdrant (~34k), pgvector (~22k), Weaviate (~17k), plus Chroma for local prototyping.
4. Browser and actions
browser-use (~107k) drives a real browser; Playwright MCP (~5.6k) exposes browser control as tools; Skyvern targets form-heavy workflows.
5. MCP
modelcontextprotocol/servers (~89k), awesome-mcp-servers (~92k), and the Python SDK (~24k). MCP is how agents connect to outside systems without rewriting tool code per model.
6. Orchestration platforms
n8n (~199k), Langflow (~153k), Dify (~151k), Open WebUI (~147k). Visual tooling for business users, code frameworks for logic under version control.
7. Observability
Langfuse (~28k) records inputs, outputs, latency, and token cost per call; Phoenix (~14k) bundles tracing with evaluation; Helicone (~8k) is the lightest to add.
Minimal stack and adoption order
LangGraph + mem0 + Qdrant + Langfuse + one MCP server. Ship one capability with framework plus retrieval first, add memory when the agent forgets, add observability before launch, and add MCP when internal systems must become tools.
FAQ
All six layers? No - fill gaps as they appear.
Cheapest setup? The stack above with small models for intermediate steps.
Chinese models? Almost every tool only needs an OpenAI-compatible endpoint.
Is MCP the standard? It is the most widely supported tool protocol today; the abstraction will outlive any implementation.
Stars and protocol details change - check each repository.
❓ FAQ
What is the cheapest stack?
All of the above are open-source and self-hostable on a single 4-core server.
Do I need all six layers?
No. Start with framework + retrieval, add memory when your agent loses context, add observability before you deploy.
Written by our editorial team; tools listed here are tested or verified against public sources. Links point to official sites or GitHub repos for reference only โ no paid placements.