Openapi MCP server
Commit 16. Okt. 2026Wraps Openapi.com APIs as an MCP server so AI clients can call them, with passthrough Bearer token auth and file handling.
Auch bekannt als openapi/mcp-server, mcp-server
- Stars
- 22
- Tools
- —
- Downloads, 30 Tage
- —
- Neueste Version
- 0.3.1
- Letzter Commit
- 16. Okt. 2026
- Lizenz
- MIT
Was ist Openapi MCP?
Openapi ist ein MCP server. Wraps Openapi.com APIs as an MCP server so AI clients can call them, with passthrough Bearer token auth and file handling.
Zusammenfassung generiert aus dem README des Projekts
Über Openapi MCP server
Connecting AI clients to Openapi.com APIs is what this official Python SDK enables. It functions as a bridge that allows developers to wrap Openapi.com services into the Model Context Protocol, making them accessible to tools like VS Code Copilot. You can run it as a ready-to-use server using a single command via uvx or integrate it as a library into your own Python projects to build custom logic.
Security is handled through a passthrough mechanism where the client provides a Bearer token via an Authorization header or a query parameter. For handling large binary documents like PDFs or ZIP archives, the server includes a storage layer. This layer can be configured to save files to a local filesystem, Google Cloud Storage, or AWS S3 buckets.
Configuration is managed entirely through environment variables, allowing for easy deployment across local machines, Docker containers, or cloud environments like Kubernetes. The server supports various caching backends, including Redis and Memcached, to manage asynchronous callback results across multiple instances.
Was du damit machen kannst
- 1Accessing official company documentsAn AI client can request and download large binary files like official company PDFs or ZIP archives through the server's storage backend.
- 2Building custom API wrappersSoftware engineers can import the SDK as a library to create specialized MCP servers with unique tools and authentication layers.
Geschrieben aus dem README des Projekts. Jeder Use Case wurde gegen ein Zitat daraus geprüft.
So installierst du Openapi MCP
Formate geprüft am 10. Nov. 2026Openapi MCP in Claude Code installieren
$ claude mcp add --transport http mcp-server https://mcp.openapi.com/Openapi MCP in Cursor installieren
~/.cursor/mcp.json
{
"mcpServers": {
"mcp-server": {
"url": "https://mcp.openapi.com/"
}
}
}Openapi MCP in VS Code installieren
.vscode/mcp.json
{
"servers": {
"mcp-server": {
"type": "http",
"url": "https://mcp.openapi.com/"
}
}
}Openapi MCP in Codex installieren
~/.codex/config.toml
[mcp_servers.mcp-server] url = "https://mcp.openapi.com/"
Openapi MCP in Gemini CLI installieren
~/.gemini/settings.json
{
"mcpServers": {
"mcp-server": {
"httpUrl": "https://mcp.openapi.com/"
}
}
}Konfiguration
MCP_STORAGE_PATH
| Name | Beschreibung | Typ |
|---|---|---|
| MCP_STORAGE_PATH | — |
Automatisierte Checks
Geluften am 14. Nov. 2026Vollständiger Bericht →Noch nicht verifiziert. Ein Server benötigt einen Score von 80, keine kritischen Befunde und genug Daten für unsere Prüfung. Wie die Checks funktionieren
- Source Repository verknüpft und erreichbaropenapi/mcp-server is public and reachablegithub.com/openapi/mcp-server ↗Bestanden
- Paket nennt dieses Repository als QuelleNicht geprüft
- Build-Herkunft (Sigstore-Attestierung)Nicht geprüft
- Open Source Lizenz (OSI)MITspdx.org/licenses/MIT.html ↗Bestanden
- Commit oder Release im letzten JahrLast commit or release 2026-10-16github.com/openapi/mcp-server ↗Bestanden
- OpenSSF ScorecardNicht geprüft
- Package-Name ähnelt keinem populären PackageNicht geprüft
- Kein kritischer Hinweis im Paket oder seinen AbhängigkeitenNicht geprüft
- Keine versteckten Anweisungen an das Modell0 tool descriptions and the README read; no override phrases, credential paths or hidden charactersBestanden
- Destruktive Tools sind markiertNicht geprüft
- Remote-Endpoint beantwortet den MCP-Handshakeinitialize answered with protocol 2025-06-18 in 77 msBestanden
README
Auf GitHub ansehenOverview
Welcome to openapi-mcp-sdk, this is the official openapi.com MCP SDK.
It ships as a Python package on PyPI and
can be used in two ways:
- Ready-to-use server — run the MCP gateway directly from the package with a single command, no cloning or setup required.
- Python library — import
openapi_mcp_sdkin your own project to build a custom MCP server that wraps openapi.com APIs with your own logic, tools, or auth layer on top.
Environment variables
All configuration is done through environment variables. None are required to start the server — defaults are suitable for local use.
Core
| Variable | Default | Description |
|---|---|---|
MCP_PORT |
8080 |
HTTP port the server listens on |
MCP_ENV |
production |
Runtime environment: dev | staging | production |
MCP_BASE_URL |
http://localhost:8080 |
Public URL of this server (used to build callback and file download URLs) |
MCP_CALLBACK_URL |
$MCP_BASE_URL/callbacks |
Explicit callback URL sent to async APIs |
MCP_OPENAPI_ENV |
(empty) | Openapi environment: dev, test, sandbox (alias of test), or empty for production |
Storage
The server needs to store files for APIs that return binary documents (e.g. company official documents). The storage backend is configurable:
| Variable | Default | Description |
|---|---|---|
MCP_STORAGE_BACKEND |
local |
local | gcs | s3 |
MCP_STORAGE_PATH |
./openapi_storage |
Local path for local backend |
MCP_STORAGE_BUCKET |
(required for cloud) | Bucket name for gcs or s3 |
MCP_STORAGE_REGION |
(required for s3) | AWS region for the S3 bucket |
Cache
Used to share async callback results across multiple server instances.
| Variable | Default | Description |
|---|---|---|
MCP_CACHE_BACKEND |
none |
none (in-process dict) | memcached | redis |
MCP_CACHE_URL |
(none) | Redis connection URL (MCP_CACHE_BACKEND=redis), e.g. redis://host:6379 |
MCP_CACHE_HOST |
(none) | Memcached host (MCP_CACHE_BACKEND=memcached) |
MCP_CACHE_PORT |
11211 |
Memcached port |
See docs/env/ for per-environment configuration guides (local,
Docker, AWS, GCP, Kubernetes).
Features
- Secure proxy: Pass-through of the Bearer Token provided by the client, without direct handling of sensitive credentials.
- Extensible: Easily add new tools/APIs by creating modules in
/apis/. - MCP-compatible: Designed according to MCP protocol best practices.
- Modular: All API call logic and tool registration is centralized in
mcp_core.py.
Quick Start — run from PyPI (recommended)
No cloning, no virtual environment. Start the server in one command:
uvx openapi-mcp-sdk server
The server starts immediately at http://localhost:8080.
Other runtimes
# pipx (ephemeral run, no install needed — similar to uvx)
pipx run openapi-mcp-sdk server
# pip (installs the package, then run the command)
pip install openapi-mcp-sdk && openapi-mcp-sdk server
# pip3 on systems where python3 is the default
pip3 install openapi-mcp-sdk && openapi-mcp-sdk server
CLI reference
openapi-mcp-sdk <command>
Commands:
server Start the MCP server (HTTP/SSE, default port 8080)
ping Ping the openapi.com APIs and report latency [coming soon]
token Generate or inspect an openapi.com Bearer token [coming soon]
Running with Docker
The project ships a single compose.yml with the mcp service.
# Build and start (production-like image)
docker compose up --build
# Run in background
docker compose up --build -d
# Follow logs
docker compose logs -f mcp
The server will be accessible at http://localhost:8080.
Debug and Development
Local development (without Docker)
The fastest iteration loop runs the server directly with uvicorn:
make start
This sets PYTHONPATH=src and starts uvicorn with auto-reload disabled. The
server listens on http://0.0.0.0:8080.
To test endpoints manually:
curl -H "Authorization: Bearer YOUR_TOKEN" http://localhost:8080
The server prints request headers and parameters to stdout — no extra configuration needed.
Remote debugging with VS Code and Docker (one click)
The repository is pre-configured for a single-click debug experience. No manual steps, no editing compose files, no terminal commands.
How it works
| File | Role |
|---|---|
docker/dev/Dockerfile |
Debug image — adds debugpy, exposes port 5678 |
compose.yml |
Compose config — dev image, debug ports, bind mount |
.vscode/tasks.json |
VS Code tasks — starts/stops the container automatically |
.vscode/launch.json |
Launch config — ties everything together under F5 |
Workflow
- Set your breakpoints anywhere in
src/. - Open Run and Debug (
Ctrl+Shift+D/⌘+Shift+D). - Select
MCP: Debug (Docker)from the dropdown. - Press F5.
VS Code will:
- Build the debug image (
docker/dev/Dockerfile) if needed. - Start the container via
compose.yml(debug image + bind mount). - Wait automatically until
debugpyis accepting connections on port 5678. - Attach
debugpy— your breakpoints are now live.
When you press ⇧F5 (Stop), VS Code detaches and tears down the container.
Keep the container alive after detaching? Remove the
"postDebugTask"line from.vscode/launch.json. You can then re-attach at any time by pressing F5 again without rebuilding.
Live code reload
compose.yml bind-mounts ./src into /app/src inside the container.
Edits to files under src/ are reflected immediately — no image rebuild
required. Just send a new request and the updated code runs.
Path mapping reference
| Host | Container | |
|---|---|---|
| Source code | ./src |
/app/src |
| MCP server | localhost:8080 |
0.0.0.0:8080 |
| debugpy listener | localhost:5678 |
0.0.0.0:5678 |
Authentication
The server accepts the Bearer token in two ways:
Option A — Authorization header (recommended)
Pass the token as a standard HTTP header:
Authorization: Bearer YOUR_TOKEN
Option B — ?token= query parameter
Some MCP clients or environments do not support custom HTTP headers. In that case the token can be embedded directly in the server URL as a query parameter:
http://YOUR_SERVER_IP:8080/?token=YOUR_TOKEN
The server automatically promotes it to an Authorization: Bearer header
before the request reaches the MCP layer, so the behaviour is identical.
When to use
?token=: Claude Desktop (as of mid-2025), some browser-based MCP playgrounds, and any client whose configuration only accepts a plain URL without a headers field.
MCP Client Configuration (VS Code)
-
Generate a Bearer Token From the portal https://console.openapi.com/oauth, create a token with the required scopes.
-
Create the
.vscode/mcp.jsonfileWith header (recommended):
{ "servers": { "openapi.com": { "type": "http", "url": "http://YOUR_SERVER_IP:8080/", "headers": { "Authorization": "Bearer YOUR_TOKEN" } } } }With
?token=query parameter (for clients that do not support custom headers):{ "servers": { "openapi.com": { "type": "http", "url": "http://YOUR_SERVER_IP:8080/?token=YOUR_TOKEN" } } } -
Test the integration
- Reload VS Code.
- Open the Copilot chat and type
@workspace. - Use the tools exposed by the MCP server.
File storage
Some openapi.com APIs (e.g. visure camerali — Italian official company documents)
return large binary files (ZIP archives, PDFs) that cannot be embedded inline in the
MCP stream. The server downloads these files, stores them, and serves them via a
separate HTTP endpoint (GET /status/{id}/files/{filename}).
Default: local filesystem
By default (MCP_STORAGE_BACKEND=local) files are written to ./openapi_storage
relative to the directory from which the server is started:
./openapi_storage/
└── <request_id>/
├── document.pdf
└── attachments.zip
Override the path with the MCP_STORAGE_PATH environment variable:
export MCP_STORAGE_PATH=/var/data/openapi_storage
uvx openapi-mcp-sdk server
The directory is created automatically on first use. For the download links to work correctly,
MCP_BASE_URLmust point to the public address of the server (default:http://localhost:8080).
Cloud storage backends
MCP_STORAGE_BACKEND |
Additional variables | Notes |
|---|---|---|
local |
MCP_STORAGE_PATH (default ./openapi_storage) |
Default — local disk or mounted volume |
gcs |
MCP_STORAGE_BUCKET |
Google Cloud Storage |
s3 |
MCP_STORAGE_BUCKET, MCP_STORAGE_REGION |
AWS S3 or any S3-compatible service |
See docs/env/ for full per-environment configuration guides.
Deployment environments
| Environment | Guide |
|---|---|
| Local machine (default) | docs/env/local.md |
| Docker / Docker Compose | docs/env/docker.md |
| Google Cloud Platform | docs/env/gcp.md |
| Amazon Web Services | docs/env/aws.md |
| Kubernetes | docs/env/kubernetes.md |
Project Structure
src/openapi_mcp_sdk/main.py: FastAPI + MCP server entry point.src/openapi_mcp_sdk/mcp_core.py: MCP initialization, API call helpers, error handling.src/openapi_mcp_sdk/apis/: Python modules defining MCP tools (one per API/scope).requirements.txt: Python dependencies.compose.yml: Docker Compose (production image by default; see comments to switch to debug).docker/latest/Dockerfile: Production Docker image.docker/dev/Dockerfile: Development image withdebugpyfor VS Code remote debugging..vscode/launch.json: VS Code debug configuration (attach to debugpy).docs/: Documentation and example configurations.
Adding New Tools/APIs
- Create or modify a file in
/apis/, following this pattern:from fastmcp import Context from typing import Any from mcp_core import make_api_call, mcp @mcp.tool async def tool_name(parameters..., ctx: Context) -> Any: # ...logic... return make_api_call(ctx, "GET", url, params=params) - Restart the server to apply the changes.
Security Notes
- Never hardcode credentials or tokens in the code.
- The server expects the Bearer Token to be provided by the client via HTTP headers.
- All API calls are proxied using the token provided by the client.
Useful Resources
Contributing
Contributions are always welcome! Whether you want to report bugs, suggest new features, improve documentation, or contribute code, your help is appreciated.
See docs/contributing.md for detailed instructions on how to get started. Please make sure to follow this project's docs/code-of-conduct.md to help maintain a welcoming and collaborative environment.
Authors
Meet the project authors:
- Simone Desantis (@SimoneDesantis)
- Marco Prosperi (@MarcoProsperi)
- Francesco Bianco (@francescobianco)
- Openapi Team (@openapi-it)
Partners
Meet our partners using Openapi or contributing to this SDK:
- Blank
- Credit Safe
- Deliveroo
- Gruppo MOL
- Jakala
- Octotelematics
- OTOQI
- PWC
- QOMODO S.R.L.
- SOUNDREEF S.P.A.
License
This project is licensed under the MIT License.
The MIT License is a permissive open-source license that allows you to freely use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the software, provided that the original copyright notice and this permission notice are included in all copies or substantial portions of the software.
In short, you are free to use this SDK in your personal, academic, or commercial projects, with minimal restrictions. The project is provided "as-is", without any warranty of any kind, either expressed or implied, including but not limited to the warranties of merchantability, fitness for a particular purpose, and non-infringement.
For more details, see the full license text at the MIT License page.
Openapi MCP Fragen
Wie füge ich Openapi MCP zu Claude Code hinzu?
Führe claude mcp add --transport http mcp-server https://mcp.openapi.com/ in deinem Terminal aus. Füge --scope user hinzu, damit der Server in jedem Projekt verfügbar ist.
Wie füge ich Openapi MCP zu Cursor hinzu?
Öffne ~/.cursor/mcp.json und füge den Server unter mcpServers hinzu, mit der Cursor-Config aus dem Installationsabschnitt oben.
How do I run the Openapi MCP server?
Run it directly from PyPI with a single command using uvx, or via pipx, pip, or Docker Compose.
How does the Openapi MCP server authenticate?
It accepts a Bearer token either as a standard Authorization header or as a ?token= query parameter, which is promoted to a header automatically.
Which clients lack header support and need the token query parameter?
Claude Desktop as of mid-2025 and some browser-based MCP playgrounds only accept a plain URL, so use ?token= there.
How are large binary files returned by APIs handled?
The server downloads them, stores them in local disk, GCS, or S3, and serves them via a separate HTTP endpoint.
Can I build a custom MCP server with this package?
Yes, openapi_mcp_sdk can be imported as a Python library to build a custom MCP server with your own tools and auth layer.