Kourier

Kourier MCP Server

Remote Model Context Protocol (MCP) server for Kourier — flat-rate hosted AI inference for coding agents.

Use the MCP server to discover plans, check model limits, configure your coding client, start human-approved subscription onboarding, and submit account-attributed feedback or error reports. Kourier serves DeepSeek V4.1 Flash through OpenAI-compatible and Anthropic-compatible inference APIs, with no per-token billing or request-volume quotas. Plans differ by concurrency.

https://kourier.sh/mcp

Transport: Streamable HTTP, stateless, JSON responses. Authentication: none required for information or onboarding; submit_feedback requires a Kourier API key in the connection's Authorization: Bearer header.

MCP provides service information and setup instructions, not model inference. Inference uses a subscription API key at https://api.kourier.sh/v1 (OpenAI-compatible) or https://api.kourier.sh/anthropic (Anthropic-compatible).

This repository documents the hosted service; there is no local server to install or run. The live MCP server provides eight tools: four read-only information tools, three onboarding tools, and one authenticated feedback tool. Recurring billing and agent access require explicit human approval in the browser before a dedicated inference key can be issued.

Quick start

Claude Code

claude mcp add --transport http kourier https://kourier.sh/mcp

Cursor / clients using mcpServers

Add this entry to your client's MCP configuration:

{

"mcpServers": { "kourier": { "url": "https://kourier.sh/mcp" } } }

Configuration formats vary by client; use the remote Streamable HTTP URL above where your client requests a server URL.

Once connected, ask: “What are Kourier's plans, and how do I configure my coding client to use it?”

Adding this MCP server does not change the model powering your coding client. Use get_setup_instructions to configure Kourier as the inference provider separately.

Tools

Connecting to MCP and using the information or onboarding tools requires no connection credentials. Onboarding status and completion require the private onboarding_token returned by create_checkout_link. Only submit_feedback requires an API-key bearer header. Keep tokens and keys out of logs, public messages, report content, and repositories.

For example, call get_setup_instructions with:

{

"harness": "claude-code" }

The server also exposes the `kourier-setup` Markdown resource and the `configure_kourier` prompt, which accepts the same optional harness value.

Feedback and error reports

Use `submit_feedback` to send product feedback or report a Kourier error. Reports are attributed to the account that owns the API key.

Authentication and privacy

• Configure the MCP connection with Authorization: Bearer . Store the key in your client's private credential settings; never add a real key to repository configuration.
• The key must be valid, unexpired, and not revoked, and the account must not be banned. An active inference subscription is not required to report a problem using an otherwise valid key.
• The public information and onboarding tools remain available without this header. For clients with no custom-header support, those tools still work, but feedback requires a header-capable client.
• Send only information the user wants to share. Exclude API keys, onboarding tokens, passwords, personal data, and private code or logs. A sanitized Kourier request ID can help investigate an error.

Arguments

NUL bytes, blank supplied strings, extra fields, and credential arguments are not accepted.

Example

The following command saves a feedback report when run with a valid key. Replace the sample message with feedback the user has approved, and set KOURIER_API_KEY privately in your environment before running it:

curl https://kourier.sh/mcp \

-H "Content-Type: application/json" \ -H "Accept: application/json, text/event-stream" \ -H "Authorization: Bearer $KOURIER_API_KEY" \ --data '{ "jsonrpc": "2.0", "id": 1, "method": "tools/call", "params": { "name": "submit_feedback", "arguments": { "kind": "feedback", "message": "Please add more coding-client setup examples to the documentation.", "client_name": "curl" } } }'

Successful submissions return a receipt with id and created_at in result.structuredContent; the same receipt is included as JSON text in result.content. Tool failures set result.isError to true and return error and error_description.

• `unauthorized` — configure a valid Kourier API-key bearer header.
• `invalid_arguments` — correct the submitted fields and limits above.
• `rate_limited` — wait the number of seconds in retry_after before submitting again.
• `service_unavailable` — receipt could not be confirmed. The report may already have been saved; do not blindly retry, because another submission can create a duplicate.

Hosted inference

Plans

Neither public plan has a five-hour, weekly, or monthly request or token quota. A concurrency slot becomes available when its request finishes. “No usage quotas” does not mean unlimited simultaneous requests.

Consult get_pricing or the pricing page for current terms; the MCP endpoint accepts JSON-RPC POST requests, not ordinary browser GET requests.

Model and endpoints

Use the exact model identifier; aliases are not supported. Check get_models and the model documentation for current availability and limits.

Getting connected

1. Create an account.
2. Choose Starter or Pro in Billing.
3. Create an API key in Dashboard → API Keys.
4. Call get_setup_instructions for your client and apply the returned settings.

For agent-assisted sign-in, follow Kourier's auth.md and the agent sign-in guide. Human approval happens in the browser; connecting to the public MCP server alone does not authorize billing or grant inference access.

Keep API keys in environment variables or your client's private credential store. Never commit them to a repository or paste them into public issues. The public information tools do not need your inference key.

Inference request with curl

With an active subscription and KOURIER_API_KEY set in your environment:

curl https://api.kourier.sh/v1/chat/completions \

-H "Authorization: Bearer $KOURIER_API_KEY" \ -H "Content-Type: application/json" \ --data '{ "model": "deepseek-v4.1-flash", "messages": [{"role": "user", "content": "Explain what this coding API supports."}], "max_tokens": 256 }'

Verify the MCP connection

Send an initialization request without an API key:

curl https://kourier.sh/mcp \

-H "Content-Type: application/json" \ -H "Accept: application/json, text/event-stream" \ --data '{ "jsonrpc": "2.0", "id": 1, "method": "initialize", "params": { "protocolVersion": "2025-11-25", "capabilities": {}, "clientInfo": {"name": "kourier-quickstart", "version": "1.0.0"} } }'

The response identifies the server as kourier-info and advertises tools, resources, and prompts. Supported protocol versions are 2025-11-25, 2025-06-18, and 2025-03-26.

List the available tools:

curl https://kourier.sh/mcp \

-H "Content-Type: application/json" \ -H "Accept: application/json, text/event-stream" \ --data '{"jsonrpc":"2.0","id":2,"method":"tools/list","params":{}}'

The endpoint does not offer a persistent SSE stream or MCP sessions. GET and DELETE return 405; use JSON-RPC POST. Normal MCP clients handle protocol initialization automatically.

Links

The hosted service is published in the official MCP Registry as `sh.kourier/info`. The repository includes server.json for registry metadata, glama.json for maintainer verification, and lhm.plugin.json for LobeHub publication. Directory account verification and review are separate from registry publication.

• Website: kourier.sh
• Pricing: kourier.sh/pricing
• Dashboard: app.kourier.sh
• Documentation: docs.kourier.sh/docs
• Agent sign-in: kourier.sh/auth.md
• MCP server card: kourier.sh/.well-known/mcp/server-card.json
• Setup skill: kourier-setup/SKILL.md

For documentation correction