{
  "markdown": "# snipara-examples\n\n**Reproducible examples and evidence for Snipara Project Intelligence.**\n\nThis repository keeps public integration examples, MCP discovery metadata, and\nbenchmark evidence in one place. It is intentionally small: no customer data,\nprivate runbooks, internal ranking code, or credentials are included.\n\n## Start here\n\n- [`server.json`](./server.json) describes the Hosted MCP server for the\n  official MCP Registry.\n- [`io.github.Snipara/snipara`](https://registry.modelcontextprotocol.io/v0.1/servers?search=Snipara&limit=100)\n  is the published official MCP Registry entry (`1.0.0`).\n- [`examples/hosted-mcp`](./examples/hosted-mcp) shows the primary HTTP setup\n  path without hard-coding a project or API key.\n- [`benchmarks/hosted-context-2026-06`](./benchmarks/hosted-context-2026-06)\n  contains the frozen June 19, 2026 context-selection evidence pack and its\n  committed [`results.json`](./benchmarks/hosted-context-2026-06/results.json).\n- [`proof/project-continuity-2026-07`](./proof/project-continuity-2026-07)\n  mirrors the measured summary from the public Project Intelligence proof\n  replay, with its owned-repository and negative-control limits kept explicit.\n\nVerify the benchmark evidence with only Python's standard library:\n\n```bash\npython3 benchmarks/hosted-context-2026-06/verify_results.py\n```\n\nThe verifier checks the case manifest hash, case coverage, reported means,\ndeltas, and token-reduction arithmetic. It does not claim that a live LLM run\nwill be byte-for-byte deterministic.\n\n## Hosted MCP\n\nSnipara is a Hosted MCP server for project intelligence and context\noptimization across AI agents. Each user connects a project-specific endpoint:\n\n```text\nhttps://api.snipara.com/mcp/YOUR_PROJECT_SLUG\n```\n\nAuthentication is supplied at runtime through `X-API-Key` or a bearer token.\nNever commit an API key to this repository.\n\n## What this benchmark does and does not prove\n\nThe frozen evidence pack compares focused Snipara context against the first\n32,000 tokens of the same indexed documentation corpus on 12 project-context\nquestions. It preserves case-level scores so the published aggregates can be\nindependently recomputed.\n\nIt is one hosted run per task, not a universal model-quality guarantee. Model\nbehavior and the live hosted index can change, so future reruns should publish a\nnew dated evidence pack instead of rewriting the historical result.\n\n## Additional controlled proof\n\nThe [Project Intelligence proof replay](https://www.snipara.com/proof) measures a\ndifferent question: whether continuity context changes what a coding agent finds\nbefore implementation and whether continuity-heavy scenarios pass across local,\nCodex CLI, and Claude runtimes. It is kept separate from the hosted-context\nbenchmark because it uses a Snipara-owned repository and does not yet include a\nnegative control where cold agents should already succeed.\n\n## Related projects\n\n- [`snipara-evals`](https://github.com/Snipara/snipara-evals) provides the\n  deterministic Project Intelligence scoring library and CLI.\n- [`snipara-companion`](https://github.com/Snipara/snipara-companion) provides\n  local workflow continuity and code-impact gates.\n- [`snipara-memory`](https://github.com/Snipara/snipara-memory) provides the\n  standalone open-source memory engine.\n- [Snipara documentation](https://www.snipara.com/docs) documents the Hosted\n  MCP, Project Intelligence, integrations, and security boundaries.\n\n## License\n\nApache-2.0. See [`LICENSE`](./LICENSE).\n",
  "bytes": 3497,
  "sha": "2f704e2060d638cdd7940a4fb8a50c1c838574995c0a7101392bf2a3433d89c7",
  "repo_slug": "snipara/snipara-examples",
  "fonte": "repo",
  "truncated": false,
  "api": "https://agentalog.com/api/listings/mcp_io_github_snipara_snipara_eb4b9ea1/readme"
}