run any ai model. compose agents, stack knowledge, connect tools. one api, pay per run.
claude mcp add mcp --transport http https://sh.inference.ac
~/.cursor/mcp.json
{
"mcpServers": {
"mcp": {
"url": "https://sh.inference.ac"
}
}
}
~/.codeium/windsurf/mcp_config.json
{
"mcpServers": {
"mcp": {
"serverUrl": "https://sh.inference.ac"
}
}
}
.vscode/mcp.json
{
"servers": {
"mcp": {
"type": "http",
"url": "https://sh.inference.ac"
}
}
}
cline_mcp_settings.json
{
"mcpServers": {
"mcp": {
"url": "https://sh.inference.ac"
}
}
}
claude_desktop_config.json
{
"mcpServers": {
"mcp": {
"command": "npx",
"args": [
"-y",
"mcp-remote",
"https://sh.inference.ac"
]
}
}
}
| Transport | URL | Auth headers |
|---|---|---|
| streamable-http | https://sh.inference.ac | none |
| streamable-http | https://api.inference.sh/mcp | none |
Great docs mapping to real tool behavior. Auth was painless.
Great docs mapping to real tool behavior. Auth was painless.
Fast responses, sensible error messages when inputs were off.
Fast responses, sensible error messages when inputs were off.
POST /mcp