{"id":29551194,"url":"https://github.com/bizrockman/modelcontextstandard","last_synced_at":"2025-10-06T07:15:05.194Z","repository":{"id":300597288,"uuid":"1006536028","full_name":"bizrockman/modelcontextstandard","owner":"bizrockman","description":"A minimal, protocol-agnostic standard to connect LLMs like drivers with external tools using existing APIs and formats like OpenAPI, HTTP, AS2, or CAN-Bus. No custom stacks. Just structured function calls.","archived":false,"fork":false,"pushed_at":"2025-06-29T16:23:37.000Z","size":33,"stargazers_count":0,"open_issues_count":0,"forks_count":0,"subscribers_count":0,"default_branch":"master","last_synced_at":"2025-07-18T06:30:27.596Z","etag":null,"topics":["agent-framework","ai-tools","design","driver","fastapi","function-calling","integration","llm","llmos","llms","mcp","mcs","openapi","protocol"],"latest_commit_sha":null,"homepage":"","language":"Python","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":null,"status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/bizrockman.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":null,"code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null}},"created_at":"2025-06-22T13:45:52.000Z","updated_at":"2025-06-29T16:23:40.000Z","dependencies_parsed_at":"2025-06-22T16:49:49.823Z","dependency_job_id":null,"html_url":"https://github.com/bizrockman/modelcontextstandard","commit_stats":null,"previous_names":["bizrockman/modelcontextstandard"],"tags_count":0,"template":false,"template_full_name":null,"purl":"pkg:github/bizrockman/modelcontextstandard","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/bizrockman%2Fmodelcontextstandard","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/bizrockman%2Fmodelcontextstandard/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/bizrockman%2Fmodelcontextstandard/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/bizrockman%2Fmodelcontextstandard/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/bizrockman","download_url":"https://codeload.github.com/bizrockman/modelcontextstandard/tar.gz/refs/heads/master","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/bizrockman%2Fmodelcontextstandard/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":278572783,"owners_count":26008876,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2022-07-04T15:15:14.044Z","status":"online","status_checked_at":"2025-10-06T02:00:05.630Z","response_time":65,"last_error":null,"robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":true,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["agent-framework","ai-tools","design","driver","fastapi","function-calling","integration","llm","llmos","llms","mcp","mcs","openapi","protocol"],"created_at":"2025-07-18T04:01:52.184Z","updated_at":"2025-10-06T07:15:05.160Z","avatar_url":"https://github.com/bizrockman.png","language":"Python","readme":"# Model Context Standard (MCS)\n\n**A lightweight alternative to MCP – removing the fluff.**\n\nMCS describes a simple, robust and protocol‑agnostic approach to connecting language models with external tools. Unlike MCP, which introduces a custom protocol stack with transport, initialization and messaging layers, MCS asks a simpler question: *What is technically necessary to connect LLMs to external systems?*\n\nAt the end of the day, function-calling LLMs need to connect to the environment, and MCS provides the absolute minimum required to make that happen. Without any further \"decoration\".\n\n## Core Idea\n\n```\nLLM ──► JSON call ──► Parser/Driver ──► Bridge ──► External API\n                       ▲                          │\n                       └──────── Result ◄─────────┘\n```\n\n1. **Bridge** – a transport layer (e.g. HTTP, CAN, AS2)\n2. **Spec** – a structured description of available functions (e.g. OpenAPI)\n\nThat’s all. The model emits structured calls, a driver executes them via the bridge and returns the result.\n\n---\n\n## Why MCS?\n\n* Uses existing standards instead of inventing new ones\n* Avoids the security pitfalls of custom stacks (unlike MCP)\n* Leverages familiar tooling such as Swagger, Postman, FastAPI, Express, Spring Boot\n* Enables reuse of existing APIs without refactoring\n* Supports optional autostart via containers (Docker)\n* Allows domain‑specific drivers: REST, EDI/AS2, CAN‑Bus, OPC‑UA …\n\nMCS acts like a **device‑driver model**: once a driver for a domain exists, any application (or LLM) can access that domain in a consistent way.\n\n---\n\n## What This Repo Provides\n\n* Reference driver: **REST over HTTP**\n* Working FastAPI backend \\\u003cplaceholder‑link\u003e\n* OpenAPI‑based service discovery\n* Minimal parser to extract and execute model‑generated calls\n\n### Quick Start\n\n```bash\n# clone and run demo driver\n$ git clone \u003crepo-url\u003e\n$ cd mcs\n$ docker compose up -d        # starts the FastAPI demo bridge\n```\n\nPoint your LLM (e.g. ChatGPT with Browser) to `http://localhost:8000/openapi.json` and watch it call the Fibonacci tool.\n\n---\n\n## Proof of Concept – try MCS in 2 minutes\n\nYou can verify the MCS pattern with *any* LLM that has web access by spinning up the tiny FastAPI demo included in this repo.\n\n```bash\n# clone on a VPS / cloud VM with a public DNS or IP\n$ git clone \u003crepo-url\u003e\n$ cd modelcontextstandard\n$ docker compose -f docker/quickstart/docker-compose.yml up -d  # exposes :8000 on your public host\n# optional: use a tunnel such as ngrok or cloudflared if you do not have a static IP\n```\n\n\u003e 🛠️ Tip: Platforms like **[Coolify](https://coolify.io)** or **Render** make one‑click deployment of Dockerised apps very easy.\n\nNo server handy? A public demo is **temporarily** available at:\n\n```\nhttps://mcs-quickstart.coolify.alsdienst.de\n```\n\n(as long as the endpoint is up).\n\nThe demo service is implemented in [`fastapi_server_mcs_quickstart.py`](mcs/examples/fastapi_server_mcs_quickstart.py) and exposes two endpoints:\n\n| Path                  | Purpose                                            |\n| --------------------- | -------------------------------------------------- |\n| `/openapi-html`       | serves the OpenAPI spec as HTML (LLM‑readable)     |\n| `/tools/fibonacci?n=` | returns *2 × Fibonacci(n)* to detect hallucination |\n\n### How to test with an LLM\n\n1. Ensure the demo is reachable under a **public domain** (or use the hosted URL above).\n2. Ask the LLM to fetch `/openapi-html` and construct the URL for the Fibonacci tool.\n3. In a second prompt, ask the LLM to visit that URL (e.g. `...?n=8`).\n4. A correct call returns **42**. If the model answers **21**, it hallucinated.\n\n| Model                 | Result | Notes                                                                                              |\n| --------------------- | ------ | -------------------------------------------------------------------------------------------------- |\n| ChatGPT (Browser)     | ✅      | Requires two prompts → [sample](https://chatgpt.com/share/68582042-280c-8009-8e18-d44cb72a4a28) |\n| Claude 3 (web access) | ✅      | Needs two‑step flow → [sample](https://claude.ai/share/57128a2d-22f8-440f-a09d-41018459d94f)       |\n| Gemini                | ❌      | Refuses second request                                                                             |\n| Grok                  | ❌      | Mis‑parses OpenAPI and builds wrong URL                                                            |\n| DeepSeek              | ❌      | Hallucinates, cannot target URL                                                                    |\n\n\u003e 🔍 POST requests are *not* required for this smoke test. Simple GETs confirm that the pattern works. Extending the driver to support POST + JSON is straightforward.\n\n\n---\n\n## How to Contribute\n\nWe welcome contributions that\n\n* refine the formal description of the MCS pattern\n* provide **new drivers** for other transport layers or data formats\n\nFeel free to fork, extend and build upon this codebase.\n\n\u003e **Proof‑of‑Work Notice** – This repository is shared *as is*. Pull requests and issues are welcome, but there is **no guarantee** of review, merging or long‑term maintenance. Contributions will be evaluated based on alignment with the research goals and available time.\n\n---\n\n## Limitations\n\nMCS is a pattern, **not** a full framework. For every new protocol or format you still need to supply a driver/bridge. This repo ships exactly one: REST over HTTP.\n\n---\n\n## License\n\n\\\u003cplaceholder‑license\u003e\n\n---\n\n## Contact\n\nOpen a [Discussion](placeholder‑link) or mention @\\\u003cyour‑username\u003e in an Issue/PR.\n\nIf you build something on top of MCS, we would love to hear about it!\n","funding_links":[],"categories":[],"sub_categories":[],"project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fbizrockman%2Fmodelcontextstandard","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fbizrockman%2Fmodelcontextstandard","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fbizrockman%2Fmodelcontextstandard/lists"}