A community-driven registry for Claude, Cursor, Windsurf, Cline & more. Not affiliated with Anthropic.
Are you the author? Sign in to claim
A universal MCP (Model Context Protocol) server to integrate any API with Claude Desktop using only Docker configuration
A universal MCP (Model Context Protocol) server to integrate any API with Claude Desktop using only Docker configurations. Point it at any Swagger/OpenAPI spec and it automatically generates tools that Claude can use.
Available on the Docker MCP Toolkit.
The easiest way to get started is through the Docker MCP Toolkit, which provides one-click setup with Docker Desktop.
Add to your claude_desktop_config.json:
{
"mcpServers": {
"my-api": {
"command": "docker",
"args": [
"run", "--rm", "-i", "--pull", "always",
"-e", "API_1_NAME=my-api",
"-e", "API_1_SWAGGER_URL=https://api.example.com/swagger.json",
"-e", "API_1_BASE_URL=https://api.example.com/v1",
"-e", "API_1_HEADER_AUTHORIZATION=Bearer YOUR_TOKEN",
"rflpazini/mcp-api-gateway:latest"
]
}
}
}
API_1_NAME=my-api \
API_1_SWAGGER_URL=https://api.example.com/swagger.json \
API_1_BASE_URL=https://api.example.com/v1 \
npx mcp-api-gateway
git clone https://github.com/rflpazini/mcp-api-gateway
cd mcp-api-gateway
docker build -t mcp-api-gateway .
docker run --rm -it \
-e API_1_NAME=test \
-e API_1_SWAGGER_URL=https://petstore.swagger.io/v2/swagger.json \
-e API_1_BASE_URL=https://petstore.swagger.io/v2 \
mcp-api-gateway
| Variable | Description | Required | Default |
|---|---|---|---|
API_N_NAME | Unique API name | Yes | — |
API_N_SWAGGER_URL | Swagger/OpenAPI spec URL | Yes | — |
API_N_BASE_URL | API base URL (overrides spec) | No | From spec |
API_N_HEADER_* | Custom headers (e.g., API_1_HEADER_AUTHORIZATION) | No | — |
API_N_HEADERS | JSON object with multiple headers | No | — |
For large APIs with hundreds of endpoints, these variables reduce the tool list payload sent to Claude:
| Variable | Description | Default |
|---|---|---|
API_N_TOOL_MODE | individual (one tool per endpoint) or grouped (group by resource/tag) | individual |
API_N_SCHEMA_MODE | full (complete schemas) or compact (strips large enums, moves optional params to descriptions) | full |
API_N_PATH_PREFIX | Comma-separated path prefixes to include (e.g., /api/v3/users,/api/v3/orders) | All paths |
API_N_TAGS | Comma-separated OpenAPI tags to include (e.g., Users,Orders) | All tags |
API_N_EXCLUDE_PARAMS | Comma-separated parameter names to strip from schemas (e.g., _clientRegion,_platform) | None |
| Variable | Description | Default |
|---|---|---|
API_N_TIMEOUT | Request timeout in milliseconds | 30000 |
API_N_MAX_RETRIES | Max retry attempts for 429/5xx errors (exponential backoff) | 3 |
MAX_RESPONSE_SIZE | Max response size in bytes before truncation | 102400 |
Tested with a 925-endpoint enterprise API:
| Configuration | Tools | Payload | Reduction |
|---|---|---|---|
| No optimization | 1093 | 2,269 KB | — |
SCHEMA_MODE=compact | 1093 | 633 KB | 72% |
TOOL_MODE=grouped | 53 | 184 KB | 92% |
PATH_PREFIX + compact | 94 | 50 KB | 98% |
Recommendation: Use API_N_TOOL_MODE=grouped for any API with more than 50 endpoints.
{
"mcpServers": {
"github-api": {
"command": "docker",
"args": [
"run", "--rm", "-i",
"-e", "API_1_NAME=github",
"-e", "API_1_SWAGGER_URL=https://api.github.com/swagger.json",
"-e", "API_1_HEADER_AUTHORIZATION=token ghp_xxxxxxxxxxxx",
"rflpazini/mcp-api-gateway:latest"
]
}
}
}
{
"mcpServers": {
"enterprise-api": {
"command": "docker",
"args": [
"run", "--rm", "-i",
"-e", "API_1_NAME=myapi",
"-e", "API_1_SWAGGER_URL=https://api.company.com/v3/api-docs",
"-e", "API_1_BASE_URL=https://api.company.com",
"-e", "API_1_TOOL_MODE=grouped",
"-e", "API_1_SCHEMA_MODE=compact",
"-e", "API_1_HEADER_AUTHORIZATION=Bearer your_token",
"rflpazini/mcp-api-gateway:latest"
]
}
}
}
Configure multiple APIs by incrementing the index (API_1_*, API_2_*, API_3_*):
{
"mcpServers": {
"company-apis": {
"command": "docker",
"args": [
"run", "--rm", "-i",
"-e", "API_1_NAME=users",
"-e", "API_1_SWAGGER_URL=https://api.company.com/users/swagger.json",
"-e", "API_1_HEADER_X_API_KEY=users_key_123",
"-e", "API_2_NAME=products",
"-e", "API_2_SWAGGER_URL=https://api.company.com/products/openapi.yaml",
"-e", "API_2_HEADER_AUTHORIZATION=Bearer products_token",
"-e", "API_3_NAME=orders",
"-e", "API_3_SWAGGER_URL=https://api.company.com/orders/spec.json",
"-e", "API_3_HEADERS={\"Authorization\":\"Bearer token\",\"X-Tenant\":\"company123\"}",
"rflpazini/mcp-api-gateway:latest"
]
}
}
}
Only load specific parts of a large API:
{
"mcpServers": {
"filtered-api": {
"command": "docker",
"args": [
"run", "--rm", "-i",
"-e", "API_1_NAME=myapi",
"-e", "API_1_SWAGGER_URL=https://api.company.com/swagger.json",
"-e", "API_1_PATH_PREFIX=/api/v3/users,/api/v3/orders",
"-e", "API_1_SCHEMA_MODE=compact",
"-e", "API_1_EXCLUDE_PARAMS=_clientRegion,_platform",
"rflpazini/mcp-api-gateway:latest"
]
}
}
}
Mount a local Swagger file into the container:
{
"mcpServers": {
"local-api": {
"command": "docker",
"args": [
"run", "--rm", "-i",
"-v", "/path/to/swagger.yaml:/swagger.yaml",
"-e", "API_1_NAME=local-api",
"-e", "API_1_SWAGGER_URL=file:///swagger.yaml",
"-e", "API_1_BASE_URL=http://host.docker.internal:3000",
"rflpazini/mcp-api-gateway:latest"
]
}
}
}
The server automatically persists cookies across requests. When an API returns Set-Cookie headers (e.g., after a login call), those cookies are stored and replayed on subsequent requests to the same API. No configuration needed.
When API_N_TOOL_MODE=grouped is set, endpoints are automatically grouped into resource tools:
/users endpoints become one myapi_Users tool)/pets, /store)operation parameterA built-in check_api_health tool lets Claude verify API connectivity. It pings each configured API's base URL and reports reachability, HTTP status, and response time.
Failed requests (429, 500, 502, 503, 504) are automatically retried with exponential backoff. The server honors Retry-After headers from rate-limited APIs.
The production image uses Google Distroless as the runtime base:
API_N_SWAGGER_URL)docker logs <container_id>API_N_TOOL_MODE=grouped to reduce tool countAPI_N_PATH_PREFIX or API_N_TAGS to filter endpointsAPI_N_SCHEMA_MODE=compact to reduce schema sizeAPI_N_TIMEOUT=60000PRs are welcome! Some ideas:
MIT License - see LICENSE file for details.
MCP server integration for DaVinci Resolve Studio
mcp-language-server gives MCP enabled clients access semantic tools like get definition, references, rename, and diagnos
Run Claude Code as an MCP server so any agent can delegate coding tasks to it
Browser automation using accessibility snapshots instead of screenshots