mcp-agent-managerlisted

# mcp-agent-manager ## Primary purpose Route user runtime requests to one scoped MCP and call the best matching tool. ## Runtime workflow 1. Infer domain and site from request. 2. Search local global index (CLI reads tool-index.jsonl; fallback to per-name cache if index missing): ```bash mcp-agent-manager tools search "<keywords>" [--name <candidate-name>] ``` 3. If ambiguous, ask user once. Do not guess or fan-out. 4. Start bridge: ```bash mcp-agent-manager session <mcp-name> ``` 5. If arguments unclear, request `tools.schema`. 6. Otherwise call `tools.call` directly. 7. Read `output_file` before close when present. 8. Close bridge. ## Domain routing table | User intent | MCP candidate | Search keywords | |---|---|---| | VM, instance, hypervisor, network, quota, volume | OpenStack site MCP | `instance`, `hypervisor`, `network`, `quota`, `volume` | | Metrics, logs, dashboard, alert, incident, Loki, Prometheus | Grafana viewer | `metrics`, `logs`, `alert`, `dashboard` | | Notebook, grounded research, source-backed answer | `notebooklm` | `notebook`, `source`, `research` | | Note, vault, knowledge base | `obsidian-local` | `note`, `vault`, `search` | | Workflow, execution, automation | n8n MCP | `workflow`, `execution`, `automation` | OpenStack site map: - Optional file: `~/.config/mcp-agent-manager/site-map.json` - Use generic site keys like `SITE01`, `SITE02`, `SITE03`, `SITE04` - `opsNN` inside a hostname is a topology alias, not an MCP name - Missi