martin-hausleitner/agent-skills/mcp-order

MCP Order

Turn MCP servers into ordered, callable CLIs and agent registrations. Use this for Ideon, NotebookLM, and any JSON-RPC MCP endpoint.

What “MCP Order” means here

1. Discover tools (tools/list)
2. Order them into a stable manifest (name, schema, risk, default args)
3. Call tools through one CLI
4. Register the server into agent MCP configs when requested

Helper CLI

python3 ~/.agents/skills/mcp-order/scripts/mcp_order.py --help

HTTP / SSE MCP (Ideon-style)

export MCP_URL=http://100.120.120.120:3025/api/mcp
export MCP_AUTH_HEADER="Authorization: Bearer $IDEON_API_KEY"
export MCP_ACCEPT="application/json, text/event-stream"
 
python3 ~/.agents/skills/mcp-order/scripts/mcp_order.py list
python3 ~/.agents/skills/mcp-order/scripts/mcp_order.py order --out /tmp/ideon.mcp-order.json
python3 ~/.agents/skills/mcp-order/scripts/mcp_order.py call list_projects '{}'
python3 ~/.agents/skills/mcp-order/scripts/mcp_order.py call get_project '{"projectId":"22222222-2222-4222-8222-222222222222"}'

stdio MCP

python3 ~/.agents/skills/mcp-order/scripts/mcp_order.py --transport stdio --cmd 'npx -y some-mcp-server' list

Generate ordered manifest

python3 ~/.agents/skills/mcp-order/scripts/mcp_order.py order \
  --name ideon \
  --out ~/.config/mcp-order/ideon.json

Manifest fields:

• name, transport, endpoint/command
• tools[]: name, description, inputSchema, risk, order
• defaults.headers, defaults.accept
• createdAt

Agent registration patterns

Codex / Claude style (conceptual)

Prefer existing agent setup tools when available (nlm setup, Cursor MCP UI, Codex config). When writing config manually, keep secrets outside git:

{
  "mcpServers": {
    "ideon": {
      "url": "http://100.120.120.120:3025/api/mcp",
      "headers": {
        "Authorization": "Bearer ${IDEON_API_KEY}",
        "Accept": "application/json, text/event-stream"
      }
    }
  }
}

NotebookLM reference

For NotebookLM use the dedicated skill notebooklm-mcp-cli (nlm setup add ...). Do not reinvent that server.

Authoring a new MCP server CLI wrapper

When asked to make an MCP-server CLI:

1. Identify transport: HTTP+SSE vs stdio.
2. Capture auth + required headers.
3. Run tools/list and freeze an ordered manifest.
4. Provide list, call, order, health subcommands.
5. Add a skill page that documents env vars, example calls, and safety.
6. Never embed long-lived secrets in the skill.

Scaffold:

python3 ~/.agents/skills/mcp-order/scripts/mcp_order.py scaffold \
  --name my-service \
  --url http://127.0.0.1:PORT/mcp \
  --out ~/code/my-service-mcp-cli

Safety

• Secrets only via env vars or ~/.config/... files mode 600.
• Default to read-only calls first (list, get_*, search_*).
• Destructive tools require explicit user intent.
• Log tool names + arg keys, not secret values.

Install / share

npx skills add Martin-Hausleitner/agent-skills@mcp-order -g -y

Local roots:

• ~/.agents/skills/mcp-order
• ~/.codex/skills/mcp-order
• ~/.claude/skills/mcp-order
• skills-hub: /Users/mh/code/skills-hub/skills/mcp-order