deploychan MCP

Public remote-MCP server that packages KISA's curated vibe-coding experience — knowledge, ready-made skills, leveling-up routes and recommendations — and serves it to any coding agent over one URL.

• Endpoint: https://mcp.deploychan.webcam/mcp
• Transport: Streamable HTTP · Access: public, read-only, no login, no keys
• License: MIT · everything (site, server, database) is open source

Point your agent at the endpoint and it gets search_knowledge, get_item, list_skills, get_skill, onboard, next_step, and list_recommended.

Connect

Claude Code

claude mcp add --transport http deploychan https://mcp.deploychan.webcam/mcp

Codex (~/.codex/config.toml)

[mcp_servers.deploychan]
url = "https://mcp.deploychan.webcam/mcp"
transport = "http"

Hermes / generic (mcp.json)

{
  "mcpServers": {
    "deploychan": {
      "type": "http",
      "url": "https://mcp.deploychan.webcam/mcp"
    }
  }
}

What the agent gets

Facet	Tools	What it returns
Knowledge	search_knowledge(query) · get_item(id)	Search for fragments, then pull the full guide by id.
Skills	list_skills() · get_skill(id)	Ready-made packs: steps, config, checklist.
Leveling up	onboard(goal) · next_step(step_id)	An ordered route for a goal; materials per step.
Recommended	list_recommended()	KISA's curated cross-type picks.

Self-hosting

The whole thing is open source. Run your own instance — locally or on a VPS — with your own content.

git clone https://github.com/howdeploy/deploychan_mcp
cd deploychan_mcp
docker compose up -d --build
# -> your MCP: http://localhost:8080/mcp

The server binds to loopback (127.0.0.1:8080) on purpose — put a reverse proxy (TLS + rate-limiting) in front for a public deployment.

Content

All content is hand-curated markdown with YAML frontmatter under content/:

content/skills/<id>/SKILL.md   # installable skill pack
content/knowledge/<id>.md      # a note / guide (full-text searchable)
content/routes/<id>.md         # a leveling-up route (ordered steps)
content/tools/<id>.md          # guide for installing a third-party tool
content/meta.yml               # endpoint, connect snippets, profile

ingest loads content/ into SQLite + FTS5 and generates web/catalog.json for the site:

python -m server.ingest

Search is full-text only (FTS5) — offline, no embeddings, no API keys. Everything in content/ is public; never put private data there.

Development

pip install -e ".[dev]"
python -m server.ingest      # build the DB + catalog
pytest                        # run the tests
python -m server.app          # serve locally at http://localhost:8080/mcp

Security

Read-only by design: the server serves content and runs nothing on your machine. No login, no keys, no request-history storage. Skills are shell commands and configs the agent runs on your machine — read what a skill does before you approve it, and treat third-party skills as untrusted code.