{
  "markdown": "<h1 align=\"center\">declaude</h1>\n\n<p align=\"center\">\n  <strong>Claude-English → plain English, as a service.</strong><br>\n  Strips sycophantic openers, hollow superlatives and hedging filler.<br>\n  Meaning, code and structure survive intact.\n</p>\n\n<p align=\"center\">\n  <a href=\"https://github.com/tenkenco/declaude/actions/workflows/ci.yml\"><img alt=\"CI\" src=\"https://github.com/tenkenco/declaude/actions/workflows/ci.yml/badge.svg\"></a>\n  <a href=\"https://github.com/tenkenco/declaude/actions/workflows/security.yml\"><img alt=\"Security\" src=\"https://github.com/tenkenco/declaude/actions/workflows/security.yml/badge.svg\"></a>\n  <a href=\"https://github.com/tenkenco/declaude/actions/workflows/daily.yml\"><img alt=\"Daily checks\" src=\"https://github.com/tenkenco/declaude/actions/workflows/daily.yml/badge.svg\"></a>\n  <a href=\"https://codecov.io/gh/tenkenco/declaude\"><img alt=\"Coverage\" src=\"https://codecov.io/gh/tenkenco/declaude/branch/main/graph/badge.svg\"></a>\n</p>\n\n<p align=\"center\">\n  <a href=\"https://registry.modelcontextprotocol.io\"><img alt=\"MCP registry\" src=\"https://img.shields.io/badge/MCP%20registry-io.github.tenkenco%2Fdeclaude-f97316\"></a>\n  <a href=\"https://speak-english.tenken.co\"><img alt=\"Live\" src=\"https://img.shields.io/badge/live-speak--english.tenken.co-4ade80\"></a>\n  <a href=\"LICENSE\"><img alt=\"License\" src=\"https://img.shields.io/badge/license-MIT-blue\"></a>\n  <img alt=\"Python\" src=\"https://img.shields.io/badge/python-3.12%2B-3776ab\">\n</p>\n\n---\n\n```\nBefore  \"Great question! Before I answer, let me make sure I understand what\n        you're asking. You want to know why the build is failing. Let me walk\n        through my thinking, and then I'll give you the answer.\"\n\nAfter   \"I understand you want to know why the build is failing. Here's what\n        I think is happening.\"\n\nBefore  \"One thing I didn't touch, but you won't want to leave hanging, is the\n        migration script. I'd be happy to walk you through it whenever you're\n        ready!\"\n\nAfter   \"One thing I didn't cover is the migration script. When you're ready,\n        I can go through it with you.\"\n```\n\nRuns on an open-source model (Qwen2.5-14B-AWQ) on our own L4 GPUs, so your text never\nreaches a commercial AI provider. Prompt logging is disabled at the model server: text is\nprocessed in memory and discarded, never written to disk, a database, or logs.\n\n**Live**: [speak-english.tenken.co](https://speak-english.tenken.co) ·\n[Get an API key](https://speak-english.tenken.co/signin) ·\n[Translate a document](https://speak-english.tenken.co/documents)\n\n| | Free | $5 / month |\n|---|---|---|\n| Translations | 100 / month | Unlimited |\n| Documents | 5 / month, 200 KB | 500 / month, 2 MB |\n| Card required | No | Yes |\n\n## Quick start\n\nIn Claude Code, install the plugin. It brings the skill and registers the hook:\n\n```\n/plugin marketplace add tenkenco/declaude\n/plugin install declaude@tenken\n```\n\nThen run `/declaude:setup`. It safely migrates any manual hook and configures a `dk_` key.\nClaude Code stores the key in secure plugin configuration. The hook rewrites replies as soon\nas a key exists. Set the `hook_enabled` option to false to stop it. The hook stays silent\nwithout a key, and it never blocks a session.\n\nFor the MCP server on its own, in Claude Code or any other MCP client:\n\n```bash\nclaude mcp add --transport http declaude https://speak-english.tenken.co/mcp\n```\n\nNo API key to paste: your client discovers OAuth, opens a browser sign-in, and holds the\ntoken. That is the whole setup.\n\n| Surface | Use it for | Docs |\n|---|---|---|\n| **MCP server** | Claude Code, Cursor, any MCP client. Tools: `translate`, `usage` | [skill](skills/declaude/SKILL.md) |\n| **Claude Code hook** | Rewrites replies as they render, costs zero Claude tokens. The plugin registers it | [hook/](hook/) |\n| **Documents** | Drop a `.md`/`.txt`, get it back rewritten | [web](https://speak-english.tenken.co/documents) |\n| **REST API** | `POST /v1/translate`, `/v1/documents`, `/v1/usage` | [skill](skills/declaude/SKILL.md#rest-api) |\n| **OpenAI-compatible** | `POST /v1/chat/completions` — point any OpenAI client at `/v1` | [skill](skills/declaude/SKILL.md#rest-api) |\n\nFull usage, authentication and quota behaviour live in\n**[skills/declaude/SKILL.md](skills/declaude/SKILL.md)** — installable as an agent skill, so\nyour agent can read it directly.\n\n## Architecture\n\n```\nclient ── API key / OAuth / Clerk JWT ──> Cloud Run gateway ── VPC ──> internal LB ──> vLLM GPU MIG\n                                              │                                    (Qwen2.5-14B, spot L4)\n                                              ├─> Firestore (usage, paid flags, key hashes)\n                                              └─> Stripe (payment link + signed webhooks)\n```\n\n- **gateway/** — FastAPI. Every boundary is injectable, which is why the suite runs without\n  network, GPU or cloud credentials.\n- **infra/** — Terraform, state in GCS, plan converges to zero diff. The GPU tier is\n  private-IP spot instances behind an internal L7 load balancer.\n- **hook/** — Claude Code hook client. Fails open; never blocks a session.\n- **hooks/** — `hooks.json`, the plugin registration Claude Code reads on install.\n- **CI** — lint, tests and coverage on every PR; security scans and a production smoke test\n  daily.\n\n<details>\n<summary><b>Operating notes</b></summary>\n\n- `/healthz` is reserved by Google Frontend on run.app — use `/health`.\n- Model swaps are one Terraform variable; 32B needs 2×L4 (quota bump), 14B fits one L4.\n- MIG template rollouts are deliberate (`OPPORTUNISTIC`): a proactive policy turned benign\n  template edits into surprise 15-minute outages.\n- Secrets live in Secret Manager only. API keys are stored as SHA-256 digests, so a database\n  leak yields no usable credential.\n- Prompt logging is disabled at the model server; request text is processed in memory and\n  discarded.\n- Developed test-first. Every production defect found while dogfooding became a regression\n  test before its fix.\n\n</details>\n\n## Development\n\n```bash\ncd gateway\nuv sync --dev\nuv run pytest -q          # 197 tests\nuv run ruff check .\n```\n\n## Credit\n\nGrew out of [gvzdv/claudish-to-english](https://github.com/gvzdv/claudish-to-english), the\noriginal local-Ollama hook. Licensed [MIT](LICENSE); upstream notice in\n[THIRD_PARTY_LICENSES.md](THIRD_PARTY_LICENSES.md).\n",
  "bytes": 6361,
  "sha": "f9bb9e9c42f184d8cab4205e3f28363992486857d31cfe93a25fa3760037a0d9",
  "repo_slug": "tenkenco/declaude",
  "fonte": "repo",
  "truncated": false,
  "api": "https://agentalog.com/api/listings/mcp_io_github_tenkenco_declaude_298d151d/readme"
}