Frankfurter MCP
Frankfurter is a useful API for latest currency exchange rates, historical data, or time series published by sources such as the European Central Bank. Should you have to access the Frankfurter API as tools for language model agents exposed over the Model Context Protocol (MCP), Frankfurter MCP is what you need.
Installation
If your objective is to use the tools available on this MCP server, please refer to the usage > client sub-section below.
The directory where you clone this repository will be referred to as the working directory or WD hereinafter.
Install just to manage project tasks.
Install uv. To install the project with its minimal dependencies in a virtual environment, run the just install in the WD. To install all non-essential dependencies (which are required for developing and testing), run just install-all instead.
Environment variables
Following is a list of environment variables that can be used to configure the application. A template of environment variables is provided in the file .env.template. Note that the default values listed in the table below are not always the same as those in the .env.template file.
The following environment variables can be specified, prefixed with FASTMCP_: HOST, PORT, DEBUG and LOG_LEVEL.
The underlying HTTP client also respects some environment variables, as documented in the HTTPX library. In addition, SSL_CERT_FILE and SSL_CERT_DIR can be configured to use self-signed certificates of hosted API endpoint or intermediate HTTP(S) proxy server(s).
Frankfurter MCP will cache calls to the Frankfurter API to improve performance. The cache happens with two different strategies. For API calls whose responses do not change for certain parameters, e.g., historical rate lookup, a least recently used (LRU) cache is used. For API calls whose responses do change, e.g., latest rate lookup, a time-to-live (TTL) cache is used with a default time-to-live set to 15 minutes. The cache parameters can be adjusted using the environment variables, see below.
| Variable | [Default value] and description |
|---|---|
LOG_LEVEL | [INFO] The level for logging. Changing this level also affects the log output of other dependent libraries that may use the same environment variable. See valid values at Python logging documentation. |
HTTPX_TIMEOUT | [5.0] The time for the underlying HTTP client to wait, in seconds, for a response from the Frankfurter API. The acceptable range of values is between 5.0 and 60.0. |
HTTPX_VERIFY_SSL | [True] This variable can be set to False to turn off SSL certificate verification, if, for instance, you are using a proxy server with a self-signed certificate. However, setting this to False is advised against: instead, use the SSL_CERT_FILE and SSL_CERT_DIR variables to properly configure self-signed certificates. |
FAST_MCP_HOST | [localhost] This variable specifies which host the MCP server must bind to unless the server transport (see below) is set to stdio. Note that running the server to bind to any IP by specifying 0.0.0.0 poses a security threat. Such a setting should only be used in demo environments. |
FAST_MCP_PORT | [8000] This variable specifies which port the MCP server must listen on unless the server transport (see below) is set to stdio. |
CORS_MIDDLEWARE_ALLOW_ORIGINS | ["localhost", "127.0.0.1"] This variable specifies Cross-Origin Resource Sharing (CORS) allowed origins for the MCP server unless the server transport (see below) is set to stdio. You must set it to "*" explicitly (and you will get a warning by doing so) if you want to test this server over an HTTP transport using the MCP inspector described below. |
MCP_SERVER_TRANSPORT | [stdio] The acceptable options are stdio, sse or streamable-http. However, in the .env.template, the default value is set to stdio. |
MCP_SERVER_INCLUDE_METADATA_IN_RESPONSE | [True] This specifies if additional metadata will be included with the MCP response from each tool call. The additional metadata, for example, will include the API URL of the Frankfurter server, amongst others, that is used to obtain the responses. |
FRANKFURTER_API_URL | [https://api.frankfurter.dev/v1] If you are self-hosting the Frankfurter API, you should change this to the API endpoint address of your deployment. |
LRU_CACHE_MAX_SIZE | [1024] The maximum size of the least recently used (LRU) cache for API calls. The acceptable range of values is between 128 and 65536. |
TTL_CACHE_MAX_SIZE | [256] The maximum size of the time-to-live (TTL) cache for API calls. The acceptable range of values is between 64 and 16384. |
TTL_CACHE_TTL_SECONDS | [900] The time limit, in seconds, of the time-to-live (TTL) cache for API calls. The acceptable range of values is between 60 and 3600. |
UVICORN_LIMIT_CONCURRENCY | [100] The maximum number of concurrent connections the server will accept. This helps prevent resource exhaustion from too many simultaneous connections. Only applies when using HTTP transports (sse or streamable-http). The acceptable range of values is between 10 and 10000. |
UVICORN_TIMEOUT_KEEP_ALIVE | [60] The timeout in seconds for keeping idle connections alive. Idle connections will be closed after this period to free up resources. Only applies when using HTTP transports (sse or streamable-http). The acceptable range of values is between 60 and 300. |
UVICORN_TIMEOUT_GRACEFUL_SHUTDOWN | [5] The timeout in seconds for graceful shutdown. The server will wait this long for active connections to complete before forcefully shutting down. Only applies when using HTTP transports (sse or streamable-http). The acceptable range of values is between 5 and 60. |
RATE_LIMIT_MAX_REQUESTS_PER_SECOND | [10.0] The maximum number of requests allowed per second using a token bucket algorithm. This implements rate limiting to prevent API abuse and ensure fair resource allocation. The acceptable range of values is between 1.0 and 10000.0. |
RATE_LIMIT_BURST_CAPACITY | [20] The burst capacity for the rate limiter, allowing short bursts of requests above the per-second limit. This provides flexibility for legitimate usage patterns while still protecting against sustained high request rates. The acceptable range of values is between 2x and 5x the RATE_LIMIT_MAX_REQUESTS_PER_SECOND value. |
REQUEST_SIZE_LIMIT_BYTES | [102400] The maximum size in bytes for HTTP request bodies (default 100KB). Requests exceeding this limit will be rejected with a 413 status code. This prevents memory exhaustion attacks from large payloads. Only applies when using HTTP transports (sse or streamable-http). The acceptable range of values is between 10240 (10KB) and 524288 (512KB). |
DOCKER_TMPFS_SIZE_MB | [100] The size in megabytes for the temporary filesystem (/tmp) when running in Docker with read-only root filesystem. This temporary storage is used for runtime file operations. Increase this value if the application requires more temporary storage for caching or processing large datasets. Only relevant when deploying with Docker Compose. |
Usage
The following sub-sections illustrate how to run the Frankfurter MCP as a server and how to access it from MCP clients.
Server
While running the server, you have the choice to use stdio transport or HTTP options (sse or the newer streamable-http).
Using default settings and MCP_SERVER_TRANSPORT set to sse or streamable-http, the MCP endpoint will be available over HTTP at http://localhost:8000/sse for the Server Sent Events (SSE) transport, or http://localhost:8000/mcp for the streamable HTTP transport.
If you want to run Frankfurter MCP with stdio transport and the default parameters, execute the commands below without using the .env.template file.
Server with uv
Optional: Copy the .env.template file to a .env file in the WD, to modify the aforementioned environment variables, if you want to use anything other than the default settings. Or, on your shell, you can export the environment variables that you wish to modify.
Run the following in the WD to start the MCP server.
uv run frankfurtermcp
Server with pip from PyPI package
Add this package from PyPI using pip in a virtual environment (possibly managed by uv, pyenv or conda) and then start the server by running the following.
Optional: Add a .env file with the contents of the .env.template file if you wish to modify the default values of the aforementioned environment variables. Or, on your shell, you can export the environment variables that you wish to modify.
pip install frankfurtermcp
python -m frankfurtermcp.server
Server using Docker
There is a Dockerfile provided in this repository, local.dockerfile, for containerising the Frankfurter MCP server. First, make a copy of the .env.template to a .env file. Then, modify the following variables in the .env file as needed.
FASTMCP_HOST: Set to0.0.0.0to allow external access to the container. This is only for local testing and is not recommended for production deployments.CORS_MIDDLEWARE_ALLOW_ORIGINS: Set to*to allow external access to the MCP server from any origin. This is needed if you want to test the server using the MCP Inspector over HTTP transport and is not recommended for production deployments.
To build the image, create the container and start it using Docker Compose, run the following in WD.
If you change the port to anything other than 8000 in .env, do remember to change the port number in docker-compose.yml.
Note: The minimum Docker Compose version needed is 2.24.0. You can check your version by running docker compose version. If you have an older version, please update Docker Desktop to get the latest Docker Compose. In addition, the backend must be BuildKit capable.
docker compose up --build
To run in detached mode (background), add the -d flag:
docker compose up -d --build
To stop the container:
docker compose down
To run the container and use the local Frankfurter API server, run the following command. Attach the -d flag to run in detached mode. Check the local_api.env.template file to specify the optional environment variables used by the local API server.
Note: Upon starting the first time, the local Frankfurter API may need some time to fetch updated exchange rates. For subsequent runs, the local Frankfurter API will use cached data and should be faster to start although it will still fetch the latest rates.
FRANKFURTER_API_URL=http://frankfurter_api:8080/v1 docker compose --profile local_api up --build frankfurtermcp frankfurter_api
To stop the container group created with the local_api profile, run the following command.
docker compose --profile local_api down
The docker-compose.yml file includes security hardening with read-only filesystem (where relevant), dropped capabilities and resource limits.
Note: The local API server is built using the latest code from the Frankfurter GitHub repository, hence this may be unstable. If you want to use a specific commit, change the context field for build under frankfurter_api_base in the docker-compose.yml file to point to the specific commit hash, e.g., https://github.com/lineofflight/frankfurter.git#0b6dbd80716f5abe27e8759fc548b74d35fa82b9 to use commit 0b6dbd80716f5abe27e8759fc548b74d35fa82b9.
Upon successful build and container start, the MCP server will be available over HTTP at http://localhost:8000/sse for the Server Sent Events (SSE) transport, or http://localhost:8000/mcp for the streamable HTTP transport. If you are also starting the local Frankfurter API server, the API endpoint will be available at http://localhost:8080/v1.
Cloud hosted servers
The currently available cloud hosted options are as follows.
- FastMCP Cloud: https://frankfurtermcp.fastmcp.app/mcp
- Glama.AI: https://glama.ai/mcp/servers/@anirbanbasu/frankfurtermcp
Client access
This sub-section explains ways for a client to connect and test the FrankfurterMCP server.
The official MCP visual inspector
The MCP Inspector is an official Model Context Protocol tool that can be used by developers to test and debug MCP servers. This is the most comprehensive way to explore the MCP server.
To use it, you must have Node.js installed. The best way to install and manage node as well as packages such as the MCP Inspector is to use the Node Version Manager (or, nvm). Once you have nvm installed, you can install and use the latest Long Term Release version of node by executing the following.
nvm install --lts
nvm use --lts
Following that (install and) run the MCP Inspector by executing the following in the WD.
npx @modelcontextprotocol/inspector uv run frankfurtermcp
This will create a local URL at port 6274 with an authentication token, which you can copy and browse to on your browser. Once on the MCP Inspector UI, press Connect to connect to the MCP server. Thereafter, you can explore the tools available on the server.
Claude Desktop, Visual Studio, and so on
The server entry to run with stdio transport that you can use with systems such as Claude Desktop, Visual Studio Code, and so on is as follows.
{
"command": "uv",
"args": [
"run",
"frankfurtermcp"
]
}
Instead of having frankfurtermcp as the last item in the list of args, you may need to specify the full path to the script, e.g., WD/.venv/bin/frankfurtermcp. Likewise, instead of using uv, you could also have the following JSON configuration with the path properly substituted for python3.12, for instance such as WD/.venv/bin/python3.12.
{
"command": "python3.12",
"args": [
"-m",
"frankfurtermcp.server"
]
}
List of available MCP features
FrankfurterMCP has the following MCP features.
Tools
The following table lists the names of the tools as exposed by the FrankfurterMCP server. The descriptions shown here are for documentation purposes, which may differ from the actual descriptions exposed over the model context protocol.
| Name | Description |
|---|---|
get_supported_currencies | Get a list of currencies supported by the Frankfurter API. |
get_latest_exchange_rates | Get latest exchange rates in specific currencies for a given base currency. |
convert_currency_latest | Convert an amount from one currency to another using the latest exchange rates. |
get_historical_exchange_rates | Get historical exchange rates for a specific date or date range in specific currencies for a given base currency. |
convert_currency_specific_date | Convert an amount from one currency to another using the exchange rates for a specific date. |
greet | Get a greeting from the FrankfurterMCP server. This is mostly used for internal testing. |
The required and optional arguments for each tool are not listed in the following table for brevity but are available to the MCP client over the protocol.
Contributing
Install prek. Then enable prek by running the following in the WD.
prek install
Pull requests are welcome. For major changes, please open an issue first to discuss what you would like to change.
Testing and coverage
To run the provided test cases, execute the following. Add the flag --capture=tee-sys to the command to display further console output.
uv run --group test pytest tests/
Invoke just test-coverage to run all the tests and generate a coverage report as follows. If all tests are run, the generated coverage report may look like the one below.
---------------------------------------------------------------------------------------- benchmark: 2 tests ---------------------------------------------------------------------------------------
Name (time in ms) Min Max Mean StdDev Median IQR Outliers OPS Rounds Iterations
---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
test_get_historical_exchange_rates 4.4944 (1.0) 5.1512 (1.0) 4.7919 (1.0) 0.2460 (1.0) 4.7819 (1.0) 0.3249 (1.0) 2;0 208.6840 (1.0) 5 1
test_get_latest_exchange_rates 4.7937 (1.07) 5.6976 (1.11) 5.3257 (1.11) 0.3345 (1.36) 5.4182 (1.13) 0.3575 (1.10) 2;0 187.7702 (0.90) 5 1
---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Legend:
Outliers: 1 Standard Deviation from Mean; 1.5 IQR (InterQuartile Range) from 1st Quartile and 3rd Quartile.
OPS: Operations Per Second, computed as 1 / Mean
=============================================================== 15 passed in 4.09s ===============================================================
Name Stmts Miss Cover Missing
---------------------------------------
TOTAL 265 0 100.00%
6 files skipped due to complete coverage.
Test coverage complete.
License
MIT.
Security considerations
This section documents security-related findings from vulnerability scans and provides context for deployment decisions.
Airtable vulnerability scan findings and rationale
Check for security-related findings from the Airtable vulnerability scan (search for frankfurtermcp) below, along with rationale and counter-arguments.
| Rule ID | Issue and counter arguments |
|---|---|
| MCP-R001 | Issue: Tools are registered dynamically at server startup without cryptographic signatures, immutable versioning, or integrity checks. The architecture permits hot-reload scenarios (via register_features pattern), but no signature verification or approval flow exists.<br/><br/>Counter arguments: Tools are not loaded from external sources or plugins—they are defined directly in the application source code. Integrity is ensured through version control and code review processes. Since tools are part of the application binary (not dynamically loaded plugins), cryptographic signing would add complexity without meaningful security benefit. |
| MCP-R004 | Issue: The server warns but accepts wildcard in CORS origins.<br/><br/>Counter arguments: This server is not intended to be run directly in a production environment when using HTTP transports. For deployments with stricter CORS origin control, users should use the .env.template defaults (127.0.0.1) and deploy the server behind their own reverse proxy with appropriate CORS origin controls at the reverse proxy level. |
| MCP-R013 | Issue: There is no support for HTTPS when the server binds to any IP other than 127.0.0.1.<br/><br/>Counter arguments: This server is not intended to be run directly in a production environment with HTTPS support when using HTTP transports. For deployments requiring HTTPS support, users should use the .env.template defaults (127.0.0.1) and deploy the server behind their own reverse proxy with appropriate HTTPS configuration. |
| MCP-R018 | Issue: There are no authentication or authorisation checks.<br/><br/>Counter arguments: This server is not intended to be run directly in a multi-user mode of operation when using HTTP transports. For deployments with access control, users should use the .env.template defaults (127.0.0.1) and deploy the server behind their own reverse proxy with appropriate security controls. |
Project status
The current status of the project is active as of the last update of this README. See the CHANGELOG for a detailed list of changes.