agent-toolbelt is an API toolset for AI agents, designed to provide agents with stock research capabilities and general infrastructure tools. The project is developed by marras0914, and the production API address is https://www.agenttoolbelt.live。
Purpose This toolset mainly serves AI agents that need financial data analysis and everyday task handling. It does not return raw OHLCV data; instead, it provides analysis results synthesized by an LLM, along with a series of general tools covering common auxiliary needs in agent work.
Main capabilities Stock research tools (7): real-time financial data based on Polygon.io, Finnhub, and Financial Modeling Prep, with analysis synthesized by Claude. Each stock tool returns a bullish/neutral/bearish conclusion, a thesis paragraph, key strengths, key risks, valuation interpretation, insider transaction interpretation, analyst consensus interpretation, and points to watch in the next earnings report. Each tool also returns a dataSources block containing a fetchedAt timestamp and success flags for each data source, making it easy to confirm the data sources on which the analysis is based.
General tools (20): provide common infrastructure capabilities for agents. Among them, rule-based tools are billed at $0.0001–$0.001 per call, and LLM-driven tools are billed at $0.005–$0.10 per call.
How to use It can be integrated in multiple ways:
- npm SDK: npm install agent-toolbelt, providing a typed client and LangChain tools.
- MCP: npx -y agent-toolbelt-mcp, supporting Claude Desktop and Claude Code. In Claude Desktop, it needs to be added to claude_desktop_config.json; in Claude Code, it can be installed with a single command. After installation, you can ask Claude to execute requests such as "Give me a full analysis of NVDA — thesis, earnings quality, insider activity, and whether it's cheap right now", and it will call the relevant tools in parallel.
- OpenAI GPT Actions: the OpenAPI specification is located at /openapi/openapi-gpt-actions.json.
- RapidAPI: already listed on the RapidAPI marketplace.
- MCP directories: can be found on Smithery, Glama,PulseMCP, and the MCP registry.
Agents can automatically discover all 27 tools at runtime.
Use cases Suitable for AI agents that need stock research analysis, such as obtaining comprehensive analysis, earnings quality, insider transaction activity, and valuation judgments for a stock; also suitable for agent workflows that need general tool support.
Limitations and additional notes The materials do not provide a specific list and detailed functional descriptions of the general tools. Regarding observability in production environments, the project author has also built Cordon, an MCP gateway that can be placed in front of any MCP server and logs every call, and supports writing rules to block calls that should not touch the production environment. Cordon provides real-time audit logs (tool name, parameters, response, latency), per-API-key policies (which tools each caller can use and under what conditions), and Slack-based human approval for high-risk tool calls. The free tier covers 1,000 events per month. Cordon's installation command is npx @getcordon/cli init, and its official website is getcordon.com.
