mcp-model-tool-calls

# GPT-6 Sol — MCP Tool Calls Over The Argolink Responses API Argolink-focused MCP tool calls over the Argolink Responses API for GPT-6 Sol. This repository is a self-contained integration reference: it explains the request shape, the asynchronous lifecycle where applicable, safe retries, and the failure states that matter in production. [GPT-6 Sol model page →](https://argolink.io/gh-mcp-model-tool-calls) · [Argolink API docs →](https://argolink.io/gh-unified-api-docs) The model page is authoritative for live availability, pricing, limits, and fields. This repository gives the workflow a searchable home; it does not invent a second model catalog. ## What is inside - `docs/model-contract.md`: model-specific facts checked against the current Argolink catalog. - `docs/workflow.md`: the intent behind this repository and the decision points that prevent accidental resubmission. - `examples/request.sh` and `examples/client.py`: safe, runnable request examples using `ARGOLINK_API_KEY`. - `tests/test_repository.py`: offline checks for the public contract and examples. - `ARGOLINK.md`: public-boundary rules for contributors. ## Quick start Set the key only in your shell or secret manager, then run the example: ```bash export ARGOLINK_API_KEY='your-key' ./examples/request.sh ``` The example uses model ID `gpt-6-sol` and endpoint family `/v1/responses`. The response is JSON; keep the returned request ID and redacted error body when diagnosing a failure. A timeout does not prove that a request was rejected, so do not submit a second request automatically. ## Model facts - Model ID: `gpt-6-sol` - Argolink model page: `https://argolink.io/gh-mcp-model-tool-calls` (tracked shortlink; destination is the model page) - Endpoint family: `/v1/responses` - Context: `1,050,000` tokens; max output: `128000` tokens. - Reasoning efforts: `low, medium, high, xhigh, max`; default: `medium`. - Listed token rates: input `$2` / output `$10` per million tokens; cached input `$0.2` when listed. ## Scope This repository has one clear search intent: **MCP tool calls over the Argolink Responses API**. It keeps the integration boundary small enough for a developer or an AI assistant to answer a concrete question, while linking the live Argolink model page for changes. ## Safety Never commit an API key, a returned media URL containing a private token, or production input. Examples are deliberately small and do not make a request during tests.