Three things you can do with FINOS MCP¶

FINOS MCP lets a local AI client consult financial-industry standards through read-only tools instead of guessing what a schema field, governance control or desktop intent means. Start with the problem you recognize.

The three examples below are real MCP tool calls, also tested automatically in CI. They run from a local checkout and do not require an LLM, account, paid inference, or a remote API. This is independent community software, not affiliated with or endorsed by FINOS.

Try a tool locally¶

Install uv, then:

git clone https://github.com/vardhjain/finos-mcp.git
cd finos-mcp
uv sync --all-packages

Pick just one command below. Each loads only the server needed for that example and exits when finished; it does not hold all standards in memory simultaneously. This is the source-checkout experience. To connect a published package to your AI client instead, follow the MCP client setup guide.

1. Map an AI governance control¶

You want to know which approved control covers an AI or MCP security risk. AIGF exposes the published AIR-* identifiers, the risks mitigated by each control, and references to other governance frameworks.

uv run --no-sync python examples/try_it.py aigf

The example calls get_control with id=mi-20. The result is checked for AIR-PREV-020, and the script prints the tool's own citation URI.

Try this prompt in a connected MCP client:

Use the AIGF get_control tool to retrieve mi-20. Explain what it mitigates, and cite the returned control identifier. Don't guess an ID.

Use it for: discovering documented controls and traceable governance references, not deciding whether an organization is legally compliant.

2. Understand a CDM trade type¶

You need to generate or validate a trade document but aren't sure what TradeState looks like. CDM 7.8.0 has vendored type descriptions and schemas and supports validation of Rune and legacy CDM JSON formats.

uv run --no-sync python examples/try_it.py cdm

The example calls describe_type with name=TradeState, checks the returned name and prints its cdm:// schema URI. You can follow up with validate_object for a real document; validation errors include JSON paths. For an intentionally invalid Rune event, see the recorded demo (its transcript is an older CDM 7.7.0 snapshot).

Try this prompt in your MCP client:

Use the CDM describe_type tool for TradeState. Show its fields, cardinalities and the schema resource I can cite.

Use it for: understanding schema structure and inspecting your generated documents without sending trade data to a hosted validation API.

3. Find an FDC3 action for a selected instrument¶

Your finance app has a selected instrument and you want to know which standard intent can show a price chart. The FDC3 server suggests intents for context objects instead of inventing action names.

uv run --no-sync python examples/try_it.py fdc3

The example calls suggest_intent with an fdc3.instrument context, ticker AAPL and goal show a price chart. The top suggestion is checked as ViewChart.

Try this prompt in a connected MCP client:

Use FDC3 suggest_intent for a context with type fdc3.instrument and ticker AAPL. My goal is to show a price chart. What standard intent is most appropriate?

Use it for: discovering interoperable desktop intent names and context requirements. The tool suggests actions; it doesn't launch apps or execute trades.

From examples to a real integration¶

The example inputs are in examples/try_it_cases.json. The executable script is examples/try_it.py. The regression tests invoke all three recipes through the actual local MCP client on CI.

For Claude Desktop, Claude Code, VS Code or Copilot, copy the version-pinned client configurations; the published v0.3.2 distributions are independently tested through real stdio MCP sessions on Python 3.12 and 3.14.

See the full tool reference, safety model, and detailed demo for more.