agent-finance

Agent-Finance Marketplace

A Claude Code plugin marketplace that ships the agent-finance plugin — a decision-support research harness for publicly traded companies (company profiles, financial statements, valuation, DCF, sector comps, bull/bear thesis, Excel 3-statement modeling).

This repo is not investment advice. Every forecast carries an explicit assumption: label and every datapoint carries a source: reference.

Plugins in this marketplace

Plugin	Version	Description
agent-finance	0.1.0	Market-finance research harness: 14 skills, 3 agents, Financial Datasets MCP, 3-statement Excel models with DCF and multi-scenario (Bull/Base/Bear).

Install

Prerequisite — Financial Datasets API key

Get an API key at financialdatasets.ai and export it in your shell before launching Claude Code:

export FINANCIAL_DATASETS_API_KEY="your-key-here"

On Windows PowerShell:

$env:FINANCIAL_DATASETS_API_KEY = "your-key-here"

The plugin's .mcp.json reads it via ${FINANCIAL_DATASETS_API_KEY} expansion. Without it, financial-datasets MCP calls return auth errors.

Option A — install from this marketplace

# Add the marketplace once
/plugin marketplace add <git-url-of-this-repo>

# Install the plugin
/plugin install agent-finance@agent-finance-marketplace

Option B — local development / testing

Load the plugin directly from a checkout, without publishing:

claude --plugin-dir ./plugins/agent-finance

Validate the plugin manifest and structure:

claude plugin validate ./plugins/agent-finance

Usage

Once the plugin is enabled, the orchestrator routes research requests:

> Research AAPL — full company package.
> Compare AAPL, MSFT, GOOGL on valuation.
> What's the FY-1 EBIT margin for NVDA?

Skills are exposed as /agent-finance:<skill> (e.g. /agent-finance:company-profile, /agent-finance:report-composer).

Outputs land in your project under:

• output/agent-finance/reports/ — composed Markdown deliverables.
• output/agent-finance/models/ — filled Excel 3-statement models.

Per-company cache (raw API responses, fetched filings) lives in the plugin's persistent data dir (${CLAUDE_PLUGIN_DATA}/companies/<TICKER>/) and survives plugin updates.

Excel model pipeline

The excel-financial-model skill uses fundamental_model_template_v2.xlsx (11 sheets, 86 named ranges, full DCF and Scenario & Sensitivity). The pipeline:

MCP data skills → model-input-builder → fill_model.py → validate_model.py
                                         (cell_map.json)   (9 blocking checks)

All three scenarios (Bull / Base / Bear) are filled in a single run. Base values come from MCP data; Bull/Bear are derived from explicit, configurable deltas in scenario_deltas.json. No data is invented — missing MCP fields are marked MISSING:<reason> and block delivery if critical.

Repository layout

See CLAUDE.md for the full layout, ground rules, and request flow.

Python development dependencies

The excel-financial-model and model-input-builder skills use openpyxl. To run the scripts outside the plugin runtime (e.g. for skill development):

pip install -r requirements.txt

Run the integration tests:

pytest plugins/agent-finance/skills/excel-financial-model/tests/test_pipeline.py -v

License

MIT.