safe-oas2mcp
A safe gateway to convert OpenAPI specs into MCP tools with secure defaults, risk inspection, and confirmation gates.
README
safe-oas2mcp
OpenAPI to MCP, safely.
safe-oas2mcp turns OpenAPI specs into MCP tools with secure defaults, risk inspection, confirmation gates, response limits, audit logging, and secret redaction.
It is not a plain OpenAPI-to-MCP converter. It is a Safe OpenAPI to MCP Gateway for Agent developers, SaaS APIs, and internal platform teams that need to expose real business APIs without handing unsafe operations to models by default.
Quick Start
Install from a local checkout:
python -m pip install -e ".[dev]"
Or run directly with uv from this checkout:
uvx --from . safe-oas2mcp --help
Inspect an OpenAPI file:
safe-oas2mcp inspect examples/todo/openapi.yaml
safe-oas2mcp inspect examples/todo/openapi.yaml --format json
Start the MCP stdio server:
safe-oas2mcp serve examples/todo/openapi.yaml
Why
Many systems already expose OpenAPI, but most converters expose too much too quickly. Agent tool calls can create, update, delete, export, transfer, or leak sensitive data if the gateway does not apply a conservative policy.
safe-oas2mcp keeps safety decisions visible:
- GET operations are enabled by default, unless risky keywords are detected.
- POST operations require confirmation by default.
- PUT and PATCH operations require confirmation by default.
- DELETE operations are disabled by default.
- Risk and status decisions include reasons.
- Tokens and API keys are injected server-side and are not shown to the model.
- Sensitive response fields are redacted.
Inspect Your API
safe-oas2mcp inspect ./openapi.yaml
Example output:
Tool Method Path Risk Status Reasons
list_tasks GET /tasks low enabled GET is enabled by default
create_task POST /tasks medium confirm POST write operation requires confirmation by default
delete_task DELETE /tasks/{task_id} critical disabled DELETE is disabled by default
Use JSON output for CI or review:
safe-oas2mcp inspect ./openapi.yaml --format json
Use a policy config during inspect:
safe-oas2mcp inspect ./openapi.yaml --config ./safe-oas2mcp.config.yaml
Serve as MCP Server
safe-oas2mcp serve ./openapi.yaml --config ./safe-oas2mcp.config.yaml
The server uses MCP stdio transport. It registers enabled and confirm tools, skips disabled tools, validates MCP input with the generated JSON Schema, builds an HTTP request, injects configured auth headers, and returns a redacted response.
Confirm tools return a request preview instead of executing the real HTTP request.
Security Defaults
| Condition | Risk | Status |
|---|---|---|
| GET | low | enabled |
| POST | medium | confirm |
| PUT | high | confirm |
| PATCH | high | confirm |
| DELETE | critical | disabled |
| money, permission, destructive, export, bulk, token, secret keywords | high or critical | confirm or disabled |
Default keyword groups include:
money: payment, refund, transfer, payout, invoice, billing, chargeidentity: user, account, customer, password, token, secret, key, email, phonepermission: admin, root, role, permission, policy, access, invitedestructive: delete, remove, destroy, revoke, disable, suspend, terminatebulk_data: export, bulk, batch, dump, report, download
Configuration
Create safe-oas2mcp.config.yaml:
base_url: https://api.example.com
auth:
type: bearer
token_env: EXAMPLE_API_TOKEN
headers:
X-Workspace-Id:
env: WORKSPACE_ID
timeout_seconds: 30
max_response_bytes: 1000000
policy:
include:
- "GET /tasks"
exclude:
- "DELETE /*"
overrides:
"DELETE /tasks/{task_id}":
status: confirm
risk: high
reason: "Allow task delete preview only"
audit:
enabled: false
path: safe-oas2mcp.audit.jsonl
Supported auth modes:
auth:
type: bearer
token_env: API_TOKEN
auth:
type: api_key
key_env: API_KEY
header_name: X-API-Key
Examples
MCP Client Setup
Docker
Build locally:
docker build -t safe-oas2mcp:local .
Inspect with Docker:
docker run --rm safe-oas2mcp:local inspect examples/todo/openapi.yaml
Serve with env-injected auth:
docker run --rm -i \
-e GITHUB_TOKEN="$GITHUB_TOKEN" \
safe-oas2mcp:local \
serve examples/github-readonly/openapi.yaml --config examples/github-readonly/safe-oas2mcp.config.example.yaml
Do not bake API tokens into images.
Development
Run checks:
python -m pytest -q
python -m ruff check .
python -m mypy
Run inspect locally:
safe-oas2mcp inspect examples/todo/openapi.yaml
Run an MCP stdio server:
safe-oas2mcp serve examples/todo/openapi.yaml
Roadmap
- More OpenAPI schema coverage
- Safer confirmation flows
- CI diff command for OpenAPI changes
- Audit log export formats
- Docker image publishing
- PyPI release
Security
Read SECURITY.md before connecting production APIs. The short version: keep auth in environment variables, inspect before serving, start readonly, and use policy overrides sparingly.
License
License is not selected yet.
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.