Coming soonAccounts and the planned features are not open yet.See what is planned
Agent tooling startup
AgentsUse builds the tools AI agents use.
We are building APIs, MCP servers, and SDKs for agent runtimes. The work starts from a simple observation: a model can only act through the tools it is given, and those tools need contracts, permissions, and operating limits that hold up outside a demo.
24 tools · 19 MCP servers · 6 frameworks · 4 stacks · 34 guides · 9 harness patterns
- APIs, MCP servers, and SDKs that give agents a defined way to search, read, compute, and act.
- Through a harness that validates calls, gates consequential actions, budgets results, and records the run.
- It maps the tool layer agents use today, with setup, limits, and sources builders can check.
Live today
Live today- Public market map: source-checked tools, MCP servers, frameworks, stacks, and guides.
- Browser workspace: saved stacks, JSON and CSV export, and a watchlist on your device.
- Production harness checklist: 24 checks for validation, gates, retries, logs, and scope.
Coming soon
Coming soon- Catalog API: verified records and check dates as structured data.
- Accounts and change alerts: account sync and alerts for watched entries. Not built yet.
- Vendor listings: planned paid offers for vendors. Not open for purchase. Readers always pay $0.
The Harness
A tool call needs an operating layer.
The harness sits between a model and its tools. It checks arguments before execution, holds consequential calls for approval, sets timeouts and retry bounds, controls how much of a result re-enters context, and leaves a record a reviewer can read. AgentsUse publishes nine patterns for that layer, three long-form guides, and notes inside selected profiles.
The AgentsUse framework is in the works, not shipped. Request early access
- Argument validation before executionMalformed arguments execute and fail downstream, or the model burns turns guessing the format you wanted.
- Permission gating and human approvalThe agent sends, deletes, or spends before anyone reviews the actual call.
- Context budgeting for tool resultsOne oversized result evicts the instructions and constraints the model still needs.
- Error recovery and bounded retriesA transient failure ends the run, or a permanent failure gets retried until latency and cost spin out of control.
Market map
The tool layer, organized by job
Category
Browser Automation
Browser drivers and agent layers for opening pages, reading content, and acting through a real browser engine.
7 verified tools
Category
Web Data and Document Extraction
Fetchers and converters that turn web pages and documents into structured input for an agent or model.
4 verified tools
Category
Search and Retrieval APIs
Search indexes and content APIs that let an agent find, rank, and read web information.
3 verified tools
Category
Code Execution Sandboxes
Isolated environments where agents can run code, install packages, and keep state without touching your machine.
1 verified tool
Category
Tool Integration Platforms
Platforms that give agents authenticated access to apps and APIs through one integration layer.
3 verified tools
Category
Agent Observability and Evaluation
Tracing, logging, and evaluation platforms for reviewing agent behavior, cost, and output quality.
4 verified tools
Category
Agent Memory
Memory layers that let agents remember users, facts, and prior work across sessions.
2 verified tools
Tool profiles
What a useful profile has to answer
Purpose, setup, operating limits, maintenance signals, and alternatives, each traced to a primary source.
Browser Automation
Playwright
A browser automation framework that drives Chromium, Firefox, and WebKit with a single API for tests, scripts, and AI agents.
Checked 2026-10-29
Web Data and Document Extraction
Firecrawl
A web data API that searches, scrapes, and crawls sites into clean markdown and structured output for AI agents.
Checked 2026-10-29
Agent Memory
Mem0
Memory layer for AI agents that persists user, session, and agent context across conversations for personalized applications.
Checked 2026-10-29
Search and Retrieval APIs
Tavily
A web access API built for AI agents: search, extract, map, crawl, and research endpoints behind one key.
Checked 2026-10-29
MCP servers
Connection details before you paste a config
Transport, runtime, placeholder configuration, environment variables, exposed tools, and security scope.
stdio / sse
Playwright MCP Server
Microsoft MCP server that automates a real browser through Playwright from accessibility snapshots, without a vision model.
Checked 2026-10-29
stdio
Filesystem MCP Server
Node.js MCP server for file operations scoped to explicitly allowed directories.
Checked 2026-10-29
streamable-http / stdio
GitHub MCP Server
Official GitHub MCP server that connects AI tools to repositories, issues, pull requests, Actions, and code analysis on GitHub.
Checked 2026-10-29
streamable-http
Context7 MCP Server
Context7 MCP server that injects current, version-specific documentation and code examples into the model prompt.
Checked 2026-10-29
Frameworks
Frameworks, read through their tool behavior
Scope, setup, MCP support, and compatibility records that link to the evidence behind each pairing.
Framework
LangChain
Agent framework for Python and TypeScript that standardizes model, tool, and retrieval interfaces so teams can swap providers without rewriting the agent loop.
Checked 2026-10-29
Framework
LangGraph
Low-level Python orchestration runtime for long-running, stateful agents that need durable execution, explicit state, and human review points.
Checked 2026-10-29
Framework
CrewAI
Python framework for role-based multi-agent work, pairing autonomous Crews with event-driven Flows for controlled production workflows.
Checked 2026-10-29
Framework
LlamaIndex
Python framework for ingestion, indexing, and retrieval that grounds agents and LLM applications in private data.
Checked 2026-10-29
Framework
Haystack
Python orchestration framework for production LLM applications built from explicit, modular pipelines and agents.
Checked 2026-10-29
Framework
Microsoft Agent Framework
Microsoft's open-source framework for building, orchestrating, and deploying AI agents and multi-agent workflows in Python, .NET, and Go, built as the direct successor to AutoGen and Semantic Kernel.
Checked 2026-10-29
Reference stacks
Workflows with the control points left in
A defined job, the components that do it, the order they run in, and the points where a person approves or intervenes.
Web research and briefing
Research Briefing Stack
Reference workflow for cited briefs: Tavily handles discovery and extraction, LangGraph runs a stateful planning loop, Langfuse records traces for review. Drawn from the AgentsUse catalog as a composition pattern.
3 verified components
Browser QA and regression checking
Browser QA Stack
Reference workflow for browser regression work: Playwright runs deterministic checks, the Playwright MCP server supports agent-led failure investigation with artifacts.
3 verified components
Turning websites and documents into agent-ready data
Web Data Pipeline Stack
Reference workflow for building an agent-ready corpus: Crawl4AI extracts pages, MarkItDown normalizes documents, Mem0 retains durable facts, LangSmith traces consumption back to inputs.
4 verified components
Coding assistance with project context
Coding Assistant Stack
Reference workflow for scoped coding assistance: Filesystem and Git provide repository access, Context7 supplies current library documentation, Sequential Thinking structures multi-step analysis.
4 verified components
Guides
Operating judgment for agent work
Delegation, permissions, failure handling, evaluation, cost, and handoff, written for the person who owns the result.
Guide
10 Tasks AI Agents Handle Well (and 5 to Keep in Your Hands)
Where AI agents earn their keep in 2026 and where they still fall short. A decision grid for delegating checkable, low-cost work and keeping judgment, money, and public commitments in human hands.
2026-09-26
Guide
Comparing AI Agent Frameworks
Framework marketing converges on the same claims. The dimensions that matter for choosing: control, observability, tooling, deployment, cost, and team fit, plus a 30-day evaluation plan and the build-versus-buy math.
2026-09-29
Guide
A Checklist Before Letting an AI Agent Touch Real Work
Twelve checks before an agent touches customers, money, or public communication. Covers least-privilege access, written instructions and limits, trial runs, action logs, and rollback, calibrated by risk tier.
2026-09-26
Guide
Agent Security Risks You Must Handle
Agents read untrusted content, run code, and call APIs with your credentials. The risks that materialize in production, from prompt injection to credential leakage, and the layered mitigations that contain them.
2026-09-29
Guide
How to Choose an AI Agent for a Specific Task
A job-first framework for choosing an agent for one specific task. Write the job down, score candidates on five criteria, run a same-task test on your own inputs, and cost the winner before it touches real work.
2026-10-04
Guide
When Not to Use AI Agents
Knowing when not to deploy an agent is a core operating skill. Where agents are overkill, where they are dangerous, the hidden costs of defaulting to them, and what to use instead.
2026-09-29
AgentsUse is building the tool layer AI agents run on. The catalog stays public because it is the map we build from: what agents can call today, where the contracts are thin, and where a better tool should exist.
Know a tool, server, or framework the map is missing? Suggest it for review. Submissions enter a queue and are never published automatically.