fd-daas-mcp

A single MCP server + CLI for skill-driven data access across financial, economic, and statistical sources. Browse thousands of indicators, build entity & indicator collections, run research workflows, and share dashboards - all behind one stdio server with 181 tools across 8 groups (187 with the optional pdf extra).

fd-daas-mcp is the engine behind the DAAS skill-driven data-fetch workflow. Point any MCP client (Claude Code, etc.) at it and the tools show up as <group>_<tool> calls - e.g. daas_list_sources, alerts_create_alert_rule, research_create.

Install

pip install fd-daas-mcp

Optional extras (all lazy-loaded - a missing extra degrades to a per-feature error, never a startup crash):

pip install "fd-daas-mcp[pdf]"          # local PDF/text vector search (sqlite-vec)
pip install "fd-daas-mcp[crew]"         # CrewAI router for the leader group
pip install "fd-daas-mcp[akshare]"      # live A-share OHLCV/fundamentals
pip install "fd-daas-mcp[position]"     # CSS/xpath/json-path rule extraction
pip install "fd-daas-mcp[repl]"         # REPL history/autocomplete
pip install "fd-daas-mcp[dev]"          # pytest + mkdocs for contributors

60-second quickstart

# 1. Install
pip install fd-daas-mcp

# 2. Point your MCP client at it. For Claude Code, add to your .mcp.json:

// .mcp.json
{
  "mcpServers": {
    "fd-daas-mcp": {
      "command": "fd-daas-mcp-server",
      "args": []
    }
  }
}

# 3. Provision the database (creates daas.db + full schema + starter sources.
#    DAAS_DATABASE_URL is optional - defaults to ./daas.db or ~/.fd-daas-mcp/):
fd-daas-mcp init
fd-daas-mcp doctor      # read-only health check (path, schema, row counts)

# 4. Or use the CLI directly:
fd-daas-mcp --help                  # see every group + tool
fd-daas-mcp daas list_sources       # invoke a tool: <group> <tool> [key=value ...]
fd-daas-mcp                          # drop into the REPL (needs [repl] extra)

On server startup the schema auto-creates and you'll see the registration report:

registry: 181 tools across 8 sources (failed=0, skipped_optional=1)

(181 tools with the default install; 187 with the pdf extra installed. The pdf group is optional and skipped when sqlite-vec is absent.)

Tool groups

The CLI tree mirrors this: fd-daas-mcp <group> <tool>. Both server and CLI are generated from the same registry.build(), so they cannot drift.

Skills

The skills/ directory (in the repo, not the wheel) contains 15 Claude Code skills that turn the raw tools into guided workflows - data fetch, research orchestration, dashboard building, rule authoring, and more. To use them:

git clone https://github.com/FindDataOfficial/fd-daas-mcp.git
# Then point Claude Code at skills/ in the clone (or symlink into ~/.claude/skills/).

See CONTRIBUTING.md and the docs site for details.

Architecture

src/fd_daas_mcp/                 # the installable package
  server.py                      # FastMCP app - registers every tool as <group>_<tool>
  cli.py                         # Click CLI - auto-generated from registry.build()
  registry.py                    # AST-harvests tools from each mcp/<group>/ (per-group sys.modules isolation)
  selfcheck.py                   # offline invariants (run_invariants())
  models.py                      # vendored SQLAlchemy schema (Base + ORM for every domain)
  mcp/                           # tool-group source (shipped as package data)
    alerts/  cron/  composite/  daas/  dashboard/  leader/  pdf/  research/

The registry loads each group's server.py and harvests @app.tool-decorated functions (inline groups) or the *_tools.py modules they import (non-inline groups), with per-group sys.modules isolation so leaf modules like database.py don't collide across groups. The cron group is loaded with suppress=True to neutralize APScheduler's import-time thread while keeping its idempotent DDL.

Development

git clone https://github.com/FindDataOfficial/fd-daas-mcp.git
cd fd-daas-mcp
uv sync --extra dev               # editable install + dev deps (pytest, mkdocs, twine)
uv run pytest                     # offline test suite
uv run python -m fd_daas_mcp.selfcheck

See CONTRIBUTING.md for the full guide.

License

MIT - © FindDataOfficial.