generated: '2026-07-21'
method: derived
source: >-
  Derived from Unsloth documentation (unsloth.ai/docs, llms.txt captured
  2026-07-21) and live /.well-known/ probes. Unsloth is an open-source local
  training/inference framework, not a hosted API provider, so most hosted-API
  standards do not apply.
standards:
  - id: openai-compatible-api
    conforms: true
    evidence: >-
      Unsloth exposes locally loaded models (including GGUFs) as an
      OpenAI-compatible API endpoint via llama-server, usable from the official
      OpenAI/Anthropic SDKs and agents like Claude Code and Codex
      (unsloth.ai/docs/basics/api,
      unsloth.ai/docs/basics/inference-and-deployment/llama-server-and-openai-endpoint,
      unsloth.ai/docs/integrations/connect-python-sdk-to-unsloth).
  - id: oauth2
    conforms: false
    evidence: >-
      No OAuth 2.0 surface; the local API is protected by a generated API key
      (llama-server long API key), and no /.well-known/oauth-authorization-server
      exists on unsloth.ai (probed 404, 2026-07-21).
  - id: oidc
    conforms: false
    evidence: No /.well-known/openid-configuration on unsloth.ai (probed 404, 2026-07-21).
  - id: mcp
    conforms: partial
    evidence: >-
      Unsloth does not publish a first-party MCP server, but documents
      connecting MCP servers to locally-run models
      (unsloth.ai/docs/basics/mcp); community MCP servers exist (e.g.
      OtotaO/unsloth-mcp-server) and are not first-party.
  - id: llms-txt
    conforms: true
    evidence: >-
      https://unsloth.ai/docs/llms.txt returns 200 (GitBook-published, with
      agent query instructions); captured verbatim in llms/unsloth-llms.txt.
  - id: rfc9116-security-txt
    conforms: false
    evidence: /.well-known/security.txt returns 404 on unsloth.ai (probed 2026-07-21).
  - id: rfc9457-problem-details
    conforms: false
    evidence: No hosted API error surface; not applicable to the local framework.