A MarkTechPost tutorial describes how to build an advanced workflow around Anthropic’s financial-services repository and reproduce its skill-driven architecture in pure Python. The workflow installs required libraries, clones the repository, and programmatically maps its agents, vertical plugins, partner integrations, managed-agent cookbooks, and financial analysis skills.

The tutorial then parses the repository’s SKILL.md files into a searchable registry and constructs a reusable SkillAgent. The agent injects selected financial playbooks into the Anthropic Messages API while supporting an iterative tool-use loop for Python calculations and file generation. Using this architecture, the workflow runs a synthetic discounted cash flow valuation, generates a WACC and terminal-growth sensitivity heatmap, performs comparable-company analysis with formatted Excel output, drafts a private-equity investment committee memo, and inspects a managed-agent deployment specification without sending a live deployment request.

The setup begins by installing Python libraries, cloning Anthropic’s financial-services repository, and preparing a Google Colab runtime for execution. The Anthropic API key is retrieved from Colab secrets, environment variables, or a secure interactive prompt. The workflow then initializes the official Anthropic SDK and selects the Claude model used to power the financial-analysis tasks.

The repository structure is inspected to identify agent plugins, vertical plugins, partner integrations, managed-agent cookbooks, and available commands. MCP configuration files are located, and the external financial data connectors defined in the repository are displayed. In this context, MCP connectors are relevant because they provide a structured way for an agent workflow to discover and call external tools or data services, while the tutorial keeps the example focused on repository inspection rather than live data access. Each SKILL.md file is then parsed to extract YAML metadata and methodology, with every unique skill registered for searchable access.

The tutorial defines tools that allow Claude to execute Python calculations and save generated deliverables inside the Colab environment. It also builds a persistent Python namespace so numerical models, tables, and intermediate variables remain available across multiple agent turns. The SkillAgent class is then created, selected financial playbooks are injected into its system prompt, and the Anthropic Messages API tool-use loop is managed.

For the valuation example, the workflow provides synthetic operating assumptions and instructs the DCF skill agent to construct a five-year unlevered cash-flow valuation. The Python execution tool is used to calculate enterprise value, equity value, implied share price, and a two-dimensional sensitivity matrix. The structured sensitivity results are then extracted and visualized as a heatmap showing the relationship between WACC, terminal growth, and implied valuation.

The comparable-company analysis uses a synthetic peer set to calculate enterprise value, EV-to-revenue, EV-to-EBITDA, and price-to-earnings multiples. The agent’s structured JSON response is converted into detailed comparable-company and summary-statistics DataFrames. The analysis is then exported to a multi-sheet Excel workbook with professional header formatting and automatic column sizing.

The workflow also applies an investment-committee memo skill to a hypothetical software buyout and calculates supporting return metrics with Python. The resulting memo is saved as a Markdown deliverable, and the generated file is verified in the output directory. The tutorial then inspects a managed-agent cookbook, displays the deployment specification, and reviews the generated artifacts and possible production extensions.

According to the tutorial, the completed workflow implements a practical Colab-based approximation of Anthropic’s financial-services skill and agent framework while preserving the repository’s methodology-driven approach to financial analysis. It combines structured skill discovery, dynamic system-prompt construction, persistent Python execution, API-based tool orchestration, and automated deliverable generation in a single reusable workflow. Because the examples use synthetic assumptions and a notebook-style sandbox, the tutorial is positioned as an implementation pattern rather than a source of live market analysis or investment recommendations.

The same agent architecture is presented as supporting multiple finance use cases, including DCF valuation, trading-comps analysis, sensitivity testing, Excel reporting, and investment committee memo preparation. Possible extensions include loading additional skills, combining multiple valuation playbooks, connecting to licensed financial data providers through MCP integrations, and replacing the tutorial sandbox with a governed production runtime in Claude Code, Cowork, or Managed Agents. For production use, the relevant follow-up considerations include access controls, data licensing, auditability of generated outputs, and validation of calculations before deliverables are used in financial workflows.

The full code is available on GitHub.