self-mcp

# self-mcp Private MCP monorepo for Aaron's custom services. The first service is `wechat-digest-mcp`, which: - discovers new WeChat public-account articles from configured feeds - extracts article body text and article images - deduplicates, caches, and persists article state locally - analyzes relevance, labels, and digest summaries - generates learning candidates for new companies, technologies, and themes - supports manual approval into a dynamic rules overlay - delivers morning digests through the existing OpenClaw WeChat bot The second service is `bluesky-social-mcp`, which: - checks whether the Bluesky lane is configured - previews post length/facets before publishing - publishes reviewed posts to Bluesky through the official AT Protocol SDK ## Layout - `services/wechat-digest-mcp` MCP stdio server and the direct morning runner - `services/bluesky-social-mcp` MCP stdio server for the public Bluesky social lane - `packages/core` config loading, sqlite state, canonical URLs, delivery helpers, message splitting - `packages/extractors` RSS / HTML list discovery and WeChat article extraction - `packages/analyzers` heuristic + model-backed article analysis - `config` sources and digest rules - `data` local runtime state and caches, not committed - `scripts` deterministic runners and Windows scheduled-task install helpers ## Learning loop - `wechat_articles.analyze` can emit `ruleCandidates[]` - candidates are stored in sqlite and surfaced back through digest reminders - approvals write into `data/rules.overlay.yaml` - the tracked base rules stay unchanged - a 19:30 follow-up reminder can nudge pending approvals once per day - production digest delivery now runs twice daily at 08:45 and 18:40 Asia/Shanghai ## Development ```powershell pnpm install pnpm build pnpm test pnpm dev:wechat-digest-mcp pnpm dev:bluesky-social-mcp ``` ## Production runner ```powershell pnpm build powershell -ExecutionPolicy Bypass -File .\scripts\run-wechat-digest.ps1 -Mode morning powershell -ExecutionPolicy Bypass -File .\scripts\run-wechat-digest.ps1 -Mode followup -DryRun ``` Install the daily 08:45 morning digest task, the 18:40 evening digest task, and the 19:30 learning follow-up task: ```powershell powershell -ExecutionPolicy Bypass -File .\scripts\install-wechat-digest-task.ps1 ``` ## Config - `config/wechat_sources.yaml` source allowlist - `config/wechat_digest_rules.yaml` interest rules, labels, models, and delivery targets - `config/bluesky_social.yaml` public-lane defaults for Bluesky Relevant environment variables: - `WECHAT_DIGEST_ROOT` - `WECHAT_DIGEST_CONFIG_DIR` - `WECHAT_DIGEST_DATA_DIR` - `WECHAT_DIGEST_OPENAI_BASE_URL` - `DASHSCOPE_API_KEY` - `OPENCLAW_CLI_WRAPPER` - `BLUESKY_HANDLE` - `BLUESKY_APP_PASSWORD` - `BLUESKY_SERVICE` If `DASHSCOPE_API_KEY` is missing, the analyzer falls back to heuristic scoring only. If `BLUESKY_HANDLE` or `BLUESKY_APP_PASSWORD` is missing, the Bluesky lane stays in preview-only/unconfigured mode. ## Current configured sources - 新智元 - 机器之心 - 量子位 - 开源星探 ## Architecture See [ARCHITECTURE.md](./ARCHITECTURE.md) for the production split between `self-mcp`, OpenClaw, and the Windows scheduled task. The first configured morning-digest sources are: - 新智元 - 机器之心 - 量子位 - 开源星探