latamfx-mcp
Enables AI agents to query public Latin American FX rates and perform auditable ledger reconciliation using a multi-rule matching engine.
README
latamfx-mcp
An MCP server that exposes public LatAm FX data and an auditable reconciliation engine to AI agents (Claude Desktop, Claude Code, or any MCP client).
latamfx-mcp lets an LLM agent answer questions like "what's the blue dollar
today?", "convert 1,500 USD to ARS at the MEP rate", or "reconcile these two
ledgers and tell me what didn't match" — entirely from free, key-less public
APIs. No credentials, no client data.
It doubles as a reference implementation of a production-shaped MCP server: hexagonal architecture, typed contracts, retries with backoff, a TTL cache, contract tests against mocked HTTP, CI, Docker and a Cloud Run deployment module.
Tools
| Tool | What it does |
|---|---|
list_fx_sources |
List supported sources (oficial, blue, MEP, CCL, mayorista, cripto, tarjeta). |
get_fx_quote |
Latest buy/sell quote for a source. |
get_fx_timeseries |
Historical buy/sell series (most recent N points). |
get_fx_stats |
min / max / mean / volatility of the mid price (computed with Polars). |
convert |
Convert an amount between currencies using a source's quote (USD↔ARS). |
reconcile |
Match two ledgers with a multi-rule engine; returns matches, misses and a match rate. |
Plus a resource: fx://sources (the source catalog as text).
The reconciliation engine
reconcile is a sanitized, generic version of intercompany / bank
reconciliation engines used in real fintech work. Rules run in priority order
and each right-side entry is consumed at most once, so the output is a valid
one-to-one assignment where every match is traceable to the rule that produced
it:
exact_reference— same non-empty external reference (score 1.0).amount_date— equal amount within a day-tolerance window (score decays with the gap).fuzzy_description— equal amount + similar free-text description above a threshold.
Architecture
Hexagonal (ports & adapters): the domain and application layers know nothing about HTTP or MCP, so the engine is pure and the data source is swappable.
flowchart TD
Agent[AI agent / MCP client] -->|tools, resources| Server[server.py · FastMCP]
Server --> App[application · FxService, ReconciliationService]
App --> Domain[domain · models + reconciliation engine]
App -->|FxProvider port| Port{{ports}}
Port -.implemented by.-> Adapter[infrastructure · DolarApiProvider]
Adapter -->|httpx + retries + TTL cache| Public[(dolarapi.com / argentinadatos.com)]
src/latamfx_mcp/
├── domain/ # pure models + reconciliation engine (no I/O)
├── ports/ # FxProvider Protocol (dependency inversion)
├── application/ # use cases: FX + reconciliation
├── infrastructure/ # httpx adapter, retry policy, TTL cache
├── config.py # env-driven settings
└── server.py # FastMCP wiring (tools + resource)
See docs/architecture.md and the
ADRs for the design decisions.
Quickstart
Requires uv.
git clone https://github.com/floreskemec/latamfx-mcp.git
cd latamfx-mcp
uv sync
uv run latamfx-mcp # starts the MCP server over stdio
Use it from Claude Code
claude mcp add latamfx -- uv --directory /absolute/path/to/latamfx-mcp run latamfx-mcp
Use it from Claude Desktop
Add to claude_desktop_config.json:
{
"mcpServers": {
"latamfx": {
"command": "uv",
"args": ["--directory", "/absolute/path/to/latamfx-mcp", "run", "latamfx-mcp"]
}
}
}
Then ask Claude: "Using latamfx, convert 1500 USD to ARS at the blue rate and show me the last 7 days of the blue dollar."
Development
uv sync
uv run pytest # tests + coverage
uv run ruff check . # lint
uv run ruff format . # format
uv run mypy # static types
Configuration is read from environment variables (all optional):
| Variable | Default | Purpose |
|---|---|---|
LATAMFX_HTTP_TIMEOUT |
10.0 |
HTTP timeout (seconds). |
LATAMFX_HTTP_RETRIES |
3 |
Max attempts on transient failures. |
LATAMFX_CACHE_TTL |
60.0 |
Quote/series cache TTL (seconds). |
Deployment
A multi-stage Dockerfile builds a slim image, and
deploy/terraform contains a minimal OpenTofu/Terraform
module to run it on Google Cloud Run. See the
deploy README.
Data sources
- dolarapi.com — latest quotes.
- ArgentinaDatos — historical series.
Both are free, community-maintained public APIs. This project is not affiliated with them; please review their terms before heavy use.
License
MIT © Gonzalo Flores Kemec
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.