mcpField
M
mcphub-operator

mcphub-operator

github.com/wyattowalshAgent workflows535 installsUpdated 8w ago
Repository

About mcphub-operator

mcphub-operator is a repository-managed skill for operating MCPHub configuration. It helps maintainers choose workflow groups, check configuration readiness, and follow scoped policies for compression and ChatGPT tunnels.

Key features

  1. 01Selects MCPHub workflow groups
  2. 02Runs configuration preflight checks
  3. 03Documents compression opt-in policy
  4. 04Defines ChatGPT tunnel restrictions
  5. 05Provides repository CLI recipes

Use cases

  1. 01Choose server groups for a local harness
  2. 02Check settings before operating MCPHub
  3. 03Configure approved compression opt-in
  4. 04Review ChatGPT tunnel access policy

MCP support

The skill guides operation of MCPHub, including MCP server group selection and the policy for ChatGPT's remote MCP tunnel. It does not itself expose tools to MCP clients.

Install

Copy the command and run it in a terminal.

$npx skills add https://github.com/wyattowalsh/agents --skill mcphub-operator

After installing, type/mcphub-operatorin Claude Code to confirm it loaded.

Bundled files

References and scripts installed along with SKILL.md.

evals/
evals.json
references/
cli-cheatsheet.md
compression-opt-in.md
group-picker.md
scripts/
check.py
doctor.py
preflight.sh
scripts/asset_toolkit/
validate_hooks.py

Details

From the project README

MCPHub Operator

Repo-managed MCPHub control plane: workflow groups, bearer endpoints, optional compression, ChatGPT tunnel policy. Smart Routing stays off in tracked config.

NOT for: editing the agent-stack sync script, live skills-sync apply installs, or PostgreSQL/pgvector smart-routing setup (use maintainer OpenSpec + explicit user request).

Dispatch

$ARGUMENTSAction
(empty)Show group picker summary, preflight command, and reference index
preflight / doctorRun bash scripts/preflight.sh from skill dir; stop on fail
groups / pickerLoad references/group-picker.md
compressionLoad references/compression-opt-in.md
tunnelExplain ChatGPT tunnel group-only policy + .env.mcphub tunnel vars
cliLoad references/cli-cheatsheet.md

Operator Contract

preflight
  1. From repo root (or pass --cwd to bundled doctor), run:
bash skills/mcphub-operator/scripts/preflight.sh
  1. Stop when JSON ok is false or any check with status: fail touches settings parity, smart routing, or client profiles.
  2. Treat warn as advisory (hub not running, bearer unset locally).
groups
  1. Default local harnesses attach harness only, with per-server endpoints available but disabled by default.
  2. Treat harness as the smallest high-signal set for context-bloat control: Brave Search, DuckDuckGo Search, DDGS (metasearch/news), Context7, DeepWiki, llms.txt catalog (llms-txt-explorer list_websites only), Fetch, and package metadata. Use check_website only via opt-in workflow groups. Prefer DDGS snippets; avoid fetch_full_content unless the user needs full pages.
  3. Use daily for routine opt-in expansion; use coding, research, review, or release for scoped agentic dev flows.
  4. Escalate to capability groups (web-search, web-read, docs, repo, browser, media, notebooks, data, design, productivity, accounts, references) when the task needs that surface.
  5. Use reasoning only for hard problems; keep reasoning-lab experimental.
  6. Attach account-backed, live-browser (Chrome DevTools and Scrapling headless scrape sessions), heavy, or experimental groups only with explicit user intent.
  7. Never attach tunnel to local harness configs unless testing the remote consumer shape.
compression
  1. Tracked mcp_settings.json keeps toolResultCompression.enabled: false.
  2. Opt-in locally via .env.mcphub only after user approves; see compression reference.
  3. Disable compression when debugging structured tool JSON or investigating truncation.
tunnel
  1. ChatGPT remote MCP uses tunnel group only (https://mcp.w4w.dev/mcp/tunnel when tunnel enabled).
  2. Do not widen ChatGPT to the global /mcp route or per-server endpoints without explicit user request.
  3. Tunnel credentials stay in .env.mcphub / ~/.cloudflared/ — never commit.
cli
  1. Prefer repo recipes: just mcphub-generate, just mcphub-generate-check, just mcphub-validate, just mcphub-doctor, just mcphub-smoke.
  2. Use upstream MCPHub CLI for live hub inspection when the process is running; see CLI reference.

Maintainer Loop

just mcphub-generate
just mcphub-generate-check
just mcphub-validate
uv run python skills/mcphub-operator/scripts/check.py

Regenerate settings after editing config/mcp-registry.json groups or servers.

References

Comments

Markdown supported