Tool catalog + copy-config + informational risk badges (minimal, internal-team scope)

## Summary

Add a **tool-level catalog with copy-config and informational risk badges** so internal teams can answer "what MCP tools do we have, who can use them, and how do I connect?" from one place.

This is deliberately **minimal**. The platform targets small internal teams that trust their own agents and do not want a heavyweight gateway/policy product. We build **visibility and onboarding only** — no enforcement, no approval queues, no virtual endpoints, no new identity system.

It extends primitives that already exist (`MCPServer` inventory, live-inventory cache, the existing UI connect-config renderer, existing authz scoping) rather than introducing parallel concepts.

## Goals

- Search every tool across every server a user can access, in one view.
- See owning team/server, trust, side effect, drift, and a computed risk badge.
- Copy a ready-to-use client config (Claude / Cursor / VS Code / raw JSON) from the same flow.

## Non-goals (explicitly out of scope)

- **Enforcement** — no `block`/`approve` policy decisions, no policy-mode rename. Existing `observe`/`allow-list` modes and `allow`/`deny` decisions stay as-is. No CRD enum migration.
- **Human approval workflow** for destructive tools.
- **Virtual governed endpoints** (multi-server composition / alias routing).
- **Automatic IdP/LDAP group→team mapping.** Identity comes from the company LDAP via the existing OIDC login fronting the corporate IdP; team membership stays **manual** for the MVP. Revisit auto-mapping only when manual membership actually hurts.

---

## Scope of work

### 1. Catalog API — `GET /api/runtime/tools`

- Build the response from the **existing** server-list path and the live-inventory cache. **Do not re-probe** servers.
- Scope results with the same authz rules used by `handleRuntimeServerList` (`services/api/internal/runtimeapi/servers.go`).
- Return one row per tool:
- `tool_name`, `description`, `server_name`, `namespace`, `team_id`, `endpoint_url`
- `declared`, `live`, `drift_status` (`declared` | `ungoverned` | `missing`)
- `required_trust`, `side_effect`, `risk_level`, `labels`, `connect_config`
- Filters: query text, namespace, team, server, trust, side effect, risk, drift.
- Consider pagination / a documented cap; `connect_config` may be heavy per row (lazy on detail view if needed).

### 2. Risk badge (informational only)

- Add an **optional, additive** `riskLevel` field to `ToolConfig` in `api/v1alpha1/mcpserver_types.go`. **No enum changes.**
- Compute a default when unset:
- `read` + low trust → `low`
- `write` or medium trust → `medium`
- `destructive` or high trust → `high`
- Drift (`ungoverned` / `missing`) stays a **separate badge**, not folded into risk — risk must stay an intrinsic tool property that doesn't flip with probe state.
- Leave `pkg/policy.Authorize` and its `Decision` struct **unchanged**. Risk does not gate calls.
- Regenerate CRDs + deepcopy; update webhook defaulting and `pkg/metadata/schema.go` (+ loader tests) for the new field.
- `services/mcp-gateway` emits `risk_level` in audit events alongside existing fields (mode, decision, reason, tool, subject) so audit can be filtered by risk. **No new decisions.**

### 3. Copy-config (promote existing renderer)

- `renderServerConnectConfig(server)` already exists in `services/ui/static/app.js` with Claude Desktop / Cursor / VS Code / raw JSON tabs. **Promote it**, don't rebuild.
- Add a visible **Copy config** action beside Copy URL / Copy JSON; open a focused drawer/modal with client tabs.
- Include server URL, auth header placeholder, adapter command where needed; add copy success/failure states.

### 4. Catalog UI

- Add a first-class **Tools** view (or catalog mode in the Servers screen): dense table, search, filters, sortable columns.
- Keep server cards; make the catalog the fastest path to "what tools do we have and who can use them?"
- Reuse `serverDisplayInventory`, `mergeToolInventory`, and related inventory helpers.
- Badges: trust, side effect, risk, drift (compact and stable). Add a risk filter.

### 5. CLI

- `mcp-runtime catalog tools` with `--query`, `--team`, `--namespace`, `--server`, `--side-effect`, `--trust`, `--risk`, `--drift`, `--output table|json|yaml`.
- `mcp-runtime catalog tool --server ` for a single detail view.
- `mcp-runtime server connect-config ` with `--client claude|cursor|vscode|json`, `--namespace`, `--domain`, `--output json|text` — generate the same shape as the UI from shared helper code where practical.
- Add `--tool-risk low|medium|high` to server init/apply flows that touch tool metadata.
- Update golden/help files when help output changes.

---

## Identity note

Users come from the company's internal **LDAP** directory via the existing OIDC login (`services/api/auth/oidc.go`) fronting the corporate IdP. The MVP needs **no new identity code** — point OIDC at the company provider and use manual team membership.

⚠️ `oidc.go` currently keys users on the `email` claim only. Confirm the corporate IdP issues a usable `email` (or adjust the claim mapping) before relying on this.

---

## Acceptance criteria

- [ ] `GET /api/runtime/tools` returns scoped, filterable tool rows built from cache (no re-probe), with the same authz as server list.
- [ ] `riskLevel` is an additive `ToolConfig` field; CRDs/deepcopy/webhook/metadata regenerated; no policy enum or decision changes.
- [ ] Computed risk default works; drift is a separate badge.
- [ ] Catalog UI: searchable/filterable/sortable table with trust/side-effect/risk/drift badges + risk filter.
- [ ] Visible **Copy config** in UI reusing the existing renderer.
- [ ] CLI: `catalog tools`, `catalog tool`, `server connect-config`, `--tool-risk` flag; golden files updated.
- [ ] Gateway audit events include `risk_level`.
- [ ] Tests: API (scoping, filters, drift states, live-inventory fallback), CLI golden/help, UI (`services/ui/main_test.go` — catalog rendering, filters, empty states, drift badges).

## Validation

```bash
gofmt -s -l .
go build -o bin/mcp-runtime ./cmd/mcp-runtime
go test ./api/... ./internal/operator/... ./internal/cli/... -count=1
go test ./services/api/... -count=1
go test ./services/ui/... -count=1
go test ./services/mcp-gateway/... -count=1
# before merge:
go test ./... -count=1 -race
go vet ./...
E2E_CACHE_MODE=1 E2E_SCENARIOS=smoke-auth bash test/e2e/kind.sh
```

## Suggested milestones

- **M1 — Catalog & Config:** `GET /api/runtime/tools`, catalog UI, `mcp-runtime catalog tools`, promote Copy config, `mcp-runtime server connect-config`.
- **M2 — Risk badges:** computed `riskLevel`, badge + filter, `risk_level` in audit events.