Agent Toolbelt

What marras0914/agent-toolbelt MCP server does

The marras0914/agent-toolbelt MCP server gives compatible AI clients access to the Agent Toolbelt API. Its catalog contains 27 tools: seven for stock research and 20 for general agent tasks.

The stock tools analyze ticker symbols using live financial data from Polygon.io, Finnhub, and Financial Modeling Prep. They cover investment theses, earnings history, insider activity, valuation, bull-versus-bear cases, stock comparisons, and competitive moats. Results include source-status information with fetch timestamps and per-provider success flags.

The utility set handles common transformations and analysis tasks. Examples include extracting structured values from text, counting tokens for multiple language models, generating JSON Schema or TypeScript validators, converting CSV and Markdown, reading URL metadata, summarizing web pages, building regular expressions or cron expressions, and packing content into a context-window budget. Other tools address prompt revision, meeting notes, contract clauses, document differences, dependency CVEs, image metadata, addresses, colors, brand tokens, and mock API responses.

How it works

The MCP package runs locally through npx and uses AGENT_TOOLBELT_KEY to authenticate requests to the hosted Agent Toolbelt API. Claude Desktop can then discover and call the exposed tools through its MCP configuration. The API also provides a tool catalog endpoint for runtime discovery.

Calls are metered by the hosted service. The free tier includes 1,000 calls per month with a 10-per-minute limit. Paid options include prepaid usage and monthly plans with higher quotas and rate limits. Individual tools have different per-call prices; stock research tools cost more than basic rule-based utilities.

Setup and configuration

Install the MCP package with npx -y agent-toolbelt-mcp. In Claude Desktop, add an agent-toolbelt entry to claude_desktop_config.json, set the command to npx, and pass -y and agent-toolbelt-mcp as arguments. Provide an Agent Toolbelt key through the AGENT_TOOLBELT_KEY environment variable.

A key can be obtained by registering through the production API. The documented registration flow starts with a free account and does not require a credit card. The same API is also available through the agent-toolbelt npm SDK, while LangChain users can create 27 structured tools from the typed client.

Tools and capabilities

• Analyze a ticker with thesis, earnings, insider, valuation, moat, or bull-and-bear tools.
• Compare two or three stocks and identify strengths, concerns, and investor-fit characteristics.
• Extract entities, count tokens, generate schemas, and create mock API data.
• Convert formats, inspect URLs, summarize web pages, and compare documents.
• Build regex and cron expressions, optimize prompts, and audit npm or PyPI dependencies.

Limitations and notes

The marras0914/agent-toolbelt MCP server depends on the hosted Agent Toolbelt API and therefore requires network access and a valid API key. Financial outputs are synthesized from external data providers; the returned source-status block indicates which providers supplied data for a request. Usage is subject to the service's quotas, rate limits, and per-tool charges.

The README explicitly documents Claude Desktop and Claude Code usage. It also mentions OpenAI GPT Actions, RapidAPI, and several MCP directories, but those are separate access or discovery options rather than additional MCP client implementations described here.