@argosvix/mcp-server 0.20.0-alpha.1 → 0.22.0-alpha.1

package/dist/tools.js

@@ -130,6 +130,24 @@ const TOOL_ARG_ALLOWLIST = {
     // は scores 同梱、 run は POST (= Pro+ で startEvalRun に渡す軸)。
     list_eval_runs: ["limit"],
     get_eval_run: ["runId"],
+    compare_eval_runs: ["baselineRunId", "candidateRunId"],
+    bulk_delete_calls: ["callIds", "dryRun"],
+    export_calls: ["startTime", "endTime", "provider", "model", "limit"],
+    list_saved_views: [],
+    create_saved_view: ["name", "filter"],
+    delete_saved_view: ["id"],
+    list_audit_log: ["limit", "eventType", "targetKind", "actorUserId", "from", "to", "cursor"],
+    aggregate_calls: ["startTime", "endTime", "groupBy", "metric", "provider", "tagKey"],
+    get_percentiles: ["startTime", "endTime", "provider", "model", "metric", "groupBy"],
+    list_projects: [],
+    create_project: ["name", "slug"],
+    rename_project: ["projectId", "name", "slug"],
+    delete_project: ["projectId"],
+    // 2026-06-05 axis 4 Tier 1 = get_account_health (= AI agent が 1 call で
+    // 自社 LLM infra の健康状態 サマリを取得)。 既存 4 endpoint (= aggregate /
+    // percentiles / llm-budget / audit) を 並列 fetch + 1 narrative response に
+    // 圧縮。 window 軸のみ 受け、 backend 新 endpoint 不要 (= 純 read aggregator)。
+    get_account_health: ["window"],
     // 2026-06-02 Codex round 2 🔴 fix = idempotencyKey 必須 path (= AI agent が
     // retry した時に backend で dedup)、 client が opaque string 64 char で carry。
     run_eval: ["name", "recentCount", "label", "promptRegistryId", "idempotencyKey"],
@@ -1205,6 +1223,379 @@ export const tools = [
             },
         },
     },
+    {
+        name: "get_percentiles",
+        description: "calls の percentile metrics を 取得 (= POST /v1/query/percentiles)。 metric = 'latency' (= レイテンシ ms) or 'cost' (= USD)、 全期間 1 数値 or groupBy='day'/'hour'/'minute' で 時系列 series。 " +
+            "AI agent が 「先週の p95 latency 推移を 日次で」 narrative で carry。 D1 SQLite (= percentile_cont 不在) で window function 経由 nearest-rank 法 計算。",
+        inputSchema: {
+            type: "object",
+            additionalProperties: false,
+            properties: {
+                startTime: {
+                    type: "string",
+                    description: "範囲開始 ISO timestamp (= UTC、 省略 = 全期間)",
+                },
+                endTime: {
+                    type: "string",
+                    description: "範囲終了 ISO timestamp",
+                },
+                provider: { type: "string", description: "provider filter" },
+                model: { type: "string", description: "model filter" },
+                metric: {
+                    type: "string",
+                    description: "metric 種別、 default = 'latency'",
+                    enum: ["latency", "cost"],
+                    default: "latency",
+                },
+                groupBy: {
+                    type: "string",
+                    description: "時系列 分割 軸 (省略 = 全期間 1 数値、 'day' = 日次、 'hour' = 時間別、 'minute' = 分別)",
+                    enum: ["day", "hour", "minute"],
+                },
+            },
+        },
+    },
+    {
+        name: "list_projects",
+        description: "自 account の active projects を 一覧取得 (= GET /v1/projects、 archived 除外)。 " +
+            "v1.5 project switcher narrative で、 AI agent が 「dev / staging / prod 環境別 観測」 carry。 Pro 5 件 / Team unlimited、 Free は default のみ。",
+        inputSchema: {
+            type: "object",
+            additionalProperties: false,
+            properties: {},
+        },
+    },
+    {
+        name: "create_project",
+        description: "新規 project を 作成 (= POST /v1/projects)。 name = 表示名、 slug = URL-safe 短い識別子 (= /^[a-z][a-z0-9-]{0,31}$/)。 " +
+            "Pro 5 件上限、 Team unlimited、 Free は 不可 (= 403)。 mutation = session 認証時 Origin/Referer 強制 (= dashboard 経由前提)。",
+        inputSchema: {
+            type: "object",
+            additionalProperties: false,
+            required: ["name", "slug"],
+            properties: {
+                name: {
+                    type: "string",
+                    description: "project 表示名 (1-64 文字)",
+                    minLength: 1,
+                    maxLength: 64,
+                },
+                slug: {
+                    type: "string",
+                    description: "URL-safe 短い識別子 (= /^[a-z][a-z0-9-]{0,31}$/、 32 字以内、 先頭小文字、 hyphens 可)",
+                    pattern: "^[a-z][a-z0-9-]{0,31}$",
+                },
+            },
+        },
+    },
+    {
+        name: "rename_project",
+        description: "既存 project の name / slug を 更新 (= PATCH /v1/projects/:id)。 name と slug は どちらか一方 / 両方指定可。 slug は URL-safe 制約 (= /^[a-z][a-z0-9-]{0,31}$/)。 default project の rename は 許可。",
+        inputSchema: {
+            type: "object",
+            additionalProperties: false,
+            required: ["projectId"],
+            properties: {
+                projectId: {
+                    type: "string",
+                    description: "対象 project の id (= list_projects で 取得した UUID)",
+                    minLength: 1,
+                    maxLength: 64,
+                },
+                name: {
+                    type: "string",
+                    description: "新しい表示名 (省略時 不変)",
+                    minLength: 1,
+                    maxLength: 64,
+                },
+                slug: {
+                    type: "string",
+                    description: "新しい slug (省略時 不変、 /^[a-z][a-z0-9-]{0,31}$/)",
+                    pattern: "^[a-z][a-z0-9-]{0,31}$",
+                },
+            },
+        },
+    },
+    {
+        name: "delete_project",
+        description: "project を soft delete (= DELETE /v1/projects/:id、 archived_at 設定で 論理削除)。 default project は 削除不可 (= accounts.default_project_id 参照 整合 のため 400)。 " +
+            "archived 後 calls / alerts は そのまま (= 過去観測は keep)、 新規 record は 別 project に carry する narrative。",
+        inputSchema: {
+            type: "object",
+            additionalProperties: false,
+            required: ["projectId"],
+            properties: {
+                projectId: {
+                    type: "string",
+                    description: "削除対象 project の id (= list_projects 経由)",
+                    minLength: 1,
+                    maxLength: 64,
+                },
+            },
+        },
+    },
+    {
+        name: "get_account_health",
+        description: "自社 LLM infra の健康状態 サマリを 1 call で取得 (= axis 4 Tier 1 = 自律 AI ops)。 既存 4 endpoint (= aggregate_calls / get_percentiles / get_llm_budget / list_audit_log) を 並列 fetch して 1 narrative response に圧縮。 " +
+            "返却 = { window, totals: {calls, costUsd, errorRate}, latency: {p50, p95, p99}, budget: {used, limit, percentUsed}, recentEvents: 件数, summary: 'ok' | 'warn' | 'critical' } 形式。 " +
+            "AI agent narrative = 「今うちの LLM infra どう？」 を 1 prompt で carry。 backend 新 endpoint 不要 = 純 read aggregator。 個別 endpoint 失敗は partial で 返す (= 1 軸 timeout が summary を 止めない)。",
+        inputSchema: {
+            type: "object",
+            additionalProperties: false,
+            properties: {
+                window: {
+                    type: "string",
+                    description: "観測窓 (= '1h' / '24h' / '7d'、 default '24h')",
+                    enum: ["1h", "24h", "7d"],
+                    default: "24h",
+                },
+            },
+        },
+    },
+    {
+        name: "aggregate_calls",
+        description: "calls の 集計 cube を 取得 (= POST /v1/query/aggregate)。 groupBy (= provider / model / day / hour / minute / tag) × metric (= cost / latency / tokens / count / error_rate) で 1 call で AI agent が 「今月の cost を model 別 に集計」 narrative carry。 " +
+            "tag mode は tagKey 必須 (= alphanumeric + _ - のみ、 例: 'env' / 'feature')。 hour mode は 168h / minute mode は 60min まで (= 超過 400)。 cost = SUM(cost_usd) / latency = AVG(latency_ms) / tokens = SUM(total_tokens) / count = COUNT(*) / error_rate = error ÷ total。 " +
+            "返却 = { groups: [{key, value, count}], total: {value, count} } 形式。 軸 1 操作系 + 自律 AI ops の 分析 narrative の coverage 拡張。",
+        inputSchema: {
+            type: "object",
+            additionalProperties: false,
+            properties: {
+                startTime: {
+                    type: "string",
+                    description: "範囲開始 ISO timestamp (= UTC、 省略 = 全期間)",
+                },
+                endTime: {
+                    type: "string",
+                    description: "範囲終了 ISO timestamp (= UTC、 省略 = 現在)",
+                },
+                groupBy: {
+                    type: "string",
+                    description: "集約軸 (= 'provider' / 'model' / 'day' / 'hour' / 'minute' / 'tag')、 default = 'provider'。 hour は 168h / minute は 60min まで",
+                    enum: ["provider", "model", "day", "hour", "minute", "tag"],
+                    default: "provider",
+                },
+                metric: {
+                    type: "string",
+                    description: "metric 種別 (= 'cost' / 'latency' / 'tokens' / 'count' / 'error_rate')、 default = 'cost'",
+                    enum: ["cost", "latency", "tokens", "count", "error_rate"],
+                    default: "cost",
+                },
+                provider: {
+                    type: "string",
+                    description: "provider filter (= 'openai' / 'anthropic' 等)、 省略 = 全 provider",
+                },
+                tagKey: {
+                    type: "string",
+                    description: "groupBy='tag' の時必須。 tags JSON 内 key 名 (alphanumeric + _- のみ、 1-64 文字)",
+                    pattern: "^[A-Za-z0-9][A-Za-z0-9_-]{0,63}$",
+                },
+            },
+        },
+    },
+    {
+        name: "list_audit_log",
+        description: "Phase B audit log を 一覧 取得 (= GET /v1/audit-log)。 自 account 限定、 admin role のみ許可 (= viewer/member は 403)。 " +
+            "AI agent が 「最近の招待 / API key revoke / プロジェクト変更」 等の 操作履歴 を 自律参照する narrative (= axis 4 自律 AI ops)。 " +
+            "filter = eventType (= 'invitation.created' / 'api_key.revoked' 等) / targetKind / actorUserId / from / to。 " +
+            "cursor pagination 対応 (= nextCursor 形式 = 'created_at|id')、 max limit 200。",
+        inputSchema: {
+            type: "object",
+            additionalProperties: false,
+            properties: {
+                limit: {
+                    type: "integer",
+                    description: "返却件数 (1-200、 デフォルト 50)",
+                    minimum: 1,
+                    maximum: 200,
+                    default: 50,
+                },
+                eventType: {
+                    type: "string",
+                    description: "event_type 完全一致 filter (= 'invitation.created' / 'api_key.revoked' / 'membership.removed' 等)",
+                },
+                targetKind: {
+                    type: "string",
+                    description: "target_kind filter (= 'invitation' / 'api_key' / 'membership' 等)",
+                },
+                actorUserId: {
+                    type: "string",
+                    description: "actor_user_id filter (= 特定 user の操作のみ抽出)",
+                },
+                from: {
+                    type: "string",
+                    description: "範囲開始 ISO timestamp (= UTC)",
+                },
+                to: {
+                    type: "string",
+                    description: "範囲終了 ISO timestamp (= UTC)",
+                },
+                cursor: {
+                    type: "string",
+                    description: "ページ送り cursor (= 前 response の nextCursor を そのまま渡す、 'created_at|id' 形式)",
+                },
+            },
+        },
+    },
+    {
+        name: "list_saved_views",
+        description: "保存済 saved views 一覧を 取得 (= GET /v1/saved-views)。 saved view = /calls page で よく使う filter (startDate/endDate/provider/model/limit) の組み合わせを 名前付きで 保存したもの。 " +
+            "AI agent は 「いつもの先週の OpenAI filter で 呼び出し見せて」 narrative で carry できる。 account 単位、 max 20 件。",
+        inputSchema: {
+            type: "object",
+            additionalProperties: false,
+            properties: {},
+        },
+    },
+    {
+        name: "create_saved_view",
+        description: "新規 saved view を 作成 / 同名なら上書き (= POST /v1/saved-views)。 name は account 内一意。 filter は SavedViewFilter shape (= startDate / endDate / provider / model / limit / preset / sortBy? / sortOrder?)。 " +
+            "AI agent が 自動で よく使う filter を 名前付き保存 narrative。 例: 「直近 7 日 GPT-4 のみ」 view を 作って 後で呼ぶ。",
+        inputSchema: {
+            type: "object",
+            additionalProperties: false,
+            required: ["name", "filter"],
+            properties: {
+                name: {
+                    type: "string",
+                    description: "saved view の名前 (1-80 文字、 改行不可)。 同名で 既存なら 上書き",
+                    minLength: 1,
+                    maxLength: 80,
+                },
+                filter: {
+                    type: "object",
+                    description: "filter shape = startDate (ISO) + endDate (ISO) + provider (空可) + model (空可) + limit (number) + preset (string|null) + sortBy? + sortOrder?",
+                    required: ["startDate", "endDate", "provider", "model", "limit", "preset"],
+                    properties: {
+                        startDate: { type: "string", description: "ISO timestamp (= 範囲開始)" },
+                        endDate: { type: "string", description: "ISO timestamp (= 範囲終了)" },
+                        provider: {
+                            type: "string",
+                            description: "プロバイダー (= 'openai' / 'anthropic' / 'google' 等)、 空 = 全 provider",
+                        },
+                        model: {
+                            type: "string",
+                            description: "モデル名、 空 = 全 model",
+                        },
+                        limit: {
+                            type: "integer",
+                            description: "返却件数 cap",
+                            minimum: 1,
+                        },
+                        preset: {
+                            type: ["string", "null"],
+                            description: "preset 識別子 (= dashboard 既定 filter、 null 可)",
+                        },
+                        sortBy: { type: "string", description: "ソート対象 column" },
+                        sortOrder: {
+                            type: "string",
+                            description: "ソート方向 ('asc' / 'desc')",
+                            enum: ["asc", "desc"],
+                        },
+                    },
+                },
+            },
+        },
+    },
+    {
+        name: "delete_saved_view",
+        description: "指定 id の saved view を 削除 (= DELETE /v1/saved-views/:id)。 自 account 限定。",
+        inputSchema: {
+            type: "object",
+            additionalProperties: false,
+            required: ["id"],
+            properties: {
+                id: {
+                    type: "string",
+                    description: "削除対象の saved view id (= UUID)",
+                    minLength: 1,
+                    maxLength: 64,
+                },
+            },
+        },
+    },
+    {
+        name: "export_calls",
+        description: "calls の large batch export (= POST /v1/query/export)。 query_calls より 高 limit (= plan 別 max records: Free 1000 / Pro 50000、 config/plans.ts)、 全 plan で利用可。 " +
+            "filter 軸 = startTime / endTime / provider / model + limit。 AI agent が 「先月分の全 GPT-4 呼び出しを取り出して傾向分析して」 narrative で 1 call carry。 " +
+            "結果 format は query_calls と 同 JSON (= AI が そのまま CSV / 統計に carry 可能)。",
+        inputSchema: {
+            type: "object",
+            additionalProperties: false,
+            properties: {
+                startTime: {
+                    type: "string",
+                    description: "範囲 開始 ISO timestamp (= UTC、 省略 = 全期間)",
+                },
+                endTime: {
+                    type: "string",
+                    description: "範囲 終了 ISO timestamp (= UTC、 省略 = 現在)",
+                },
+                provider: {
+                    type: "string",
+                    description: "プロバイダー fix (= openai / anthropic / google / azure / cohere)",
+                },
+                model: {
+                    type: "string",
+                    description: "model 名 fix (= 部分一致なし、 完全一致 例: 'gpt-4o-mini')",
+                },
+                limit: {
+                    type: "integer",
+                    description: "返却件数 cap。 plan 別 max 内なら そのまま、 超過は plan max に clamp",
+                    minimum: 1,
+                },
+            },
+        },
+    },
+    {
+        name: "bulk_delete_calls",
+        description: "指定 call id 一覧 (= max 100) を 自 account 限定で 一括削除する (= POST /v1/calls/bulk-delete)。 " +
+            "AI agent が dogfood / dev test で 蓄積した garbage call の cleanup narrative に carry (= 軸 1 操作系)。 " +
+            "dryRun=true で 削除前に matched 件数を 事前確認可能。 削除は 1 SQL atomic、 audit log に bulk_deleted event を 記録。 " +
+            "FK 制約上 関連 traces / annotations / scores は ON DELETE 経由で 連鎖削除 (= 既存 schema narrative)。",
+        inputSchema: {
+            type: "object",
+            additionalProperties: false,
+            required: ["callIds"],
+            properties: {
+                callIds: {
+                    type: "array",
+                    description: "削除対象の call id 配列 (1-100 件、 各 1-128 文字)",
+                    items: { type: "string", minLength: 1, maxLength: 128 },
+                    minItems: 1,
+                    maxItems: 100,
+                },
+                dryRun: {
+                    type: "boolean",
+                    description: "true で 削除せず matched 件数のみ 返却 (= 確認 UX)",
+                    default: false,
+                },
+            },
+        },
+    },
+    {
+        name: "compare_eval_runs",
+        description: "2 つの eval run (baseline / candidate) を 比較して per-criterion mean score delta + failed count delta + verdict を 返す (= GET /v1/eval-runs/compare)。 " +
+            "AI agent は 「baseline と 比べて candidate は どう 変わったか」 を 1 call で 把握でき、 prompt 改善 効果や regress 検出 narrative に carry できる (= axis 1 操作系 + axis 4 自律 AI ops 寄与)。 " +
+            "verdict = improved / regressed / mixed / unchanged。 failed count は score <= 2 を 「failed」 で 算出。 同 account 限定。",
+        inputSchema: {
+            type: "object",
+            additionalProperties: false,
+            required: ["baselineRunId", "candidateRunId"],
+            properties: {
+                baselineRunId: {
+                    type: "integer",
+                    description: "比較元 run の id (= list_eval_runs.runs[].id)",
+                    minimum: 1,
+                },
+                candidateRunId: {
+                    type: "integer",
+                    description: "比較先 run の id (= 同上)、 baseline と 異なる必要あり",
+                    minimum: 1,
+                },
+            },
+        },
+    },
     {
         name: "run_eval",
         description: "新規 eval run を 即時実行する (= POST /v1/eval-runs)。 直近 N 件の calls × 既定 5 criteria (+ 自作 custom criteria max 8) で gpt-4o-mini に採点させる。 " +
@@ -1605,6 +1996,287 @@ export async function dispatchTool(input) {
                 }
                 return await callApi(apiBase, `/v1/eval-runs/${encodeURIComponent(runId)}`, {}, apiKey);
             }
+            case "get_percentiles": {
+                const body = {};
+                if (typeof safeArgs["startTime"] === "string")
+                    body["startTime"] = safeArgs["startTime"];
+                if (typeof safeArgs["endTime"] === "string")
+                    body["endTime"] = safeArgs["endTime"];
+                if (typeof safeArgs["provider"] === "string")
+                    body["provider"] = safeArgs["provider"];
+                if (typeof safeArgs["model"] === "string")
+                    body["model"] = safeArgs["model"];
+                if (typeof safeArgs["metric"] === "string")
+                    body["metric"] = safeArgs["metric"];
+                if (typeof safeArgs["groupBy"] === "string")
+                    body["groupBy"] = safeArgs["groupBy"];
+                return await callApi(apiBase, "/v1/query/percentiles", {}, apiKey, {
+                    method: "POST",
+                    jsonBody: body,
+                });
+            }
+            case "list_projects": {
+                return await callApi(apiBase, "/v1/projects", {}, apiKey);
+            }
+            case "create_project": {
+                const name = safeArgs["name"];
+                const slug = safeArgs["slug"];
+                if (typeof name !== "string" || name.length === 0 || name.length > 64) {
+                    return errorResponse("name required (1-64 chars)");
+                }
+                if (typeof slug !== "string" || !/^[a-z][a-z0-9-]{0,31}$/.test(slug)) {
+                    return errorResponse("slug required (lowercase alphanumeric + hyphens, max 32 chars, starts with letter)");
+                }
+                return await callApi(apiBase, "/v1/projects", {}, apiKey, {
+                    method: "POST",
+                    jsonBody: { name, slug },
+                });
+            }
+            case "rename_project": {
+                const projectId = safeArgs["projectId"];
+                if (typeof projectId !== "string" || projectId.length === 0 || projectId.length > 64) {
+                    return errorResponse("projectId required (1-64 chars)");
+                }
+                const body = {};
+                if (typeof safeArgs["name"] === "string")
+                    body["name"] = safeArgs["name"];
+                if (typeof safeArgs["slug"] === "string") {
+                    if (!/^[a-z][a-z0-9-]{0,31}$/.test(safeArgs["slug"])) {
+                        return errorResponse("slug must match /^[a-z][a-z0-9-]{0,31}$/");
+                    }
+                    body["slug"] = safeArgs["slug"];
+                }
+                if (Object.keys(body).length === 0) {
+                    return errorResponse("at least one of name / slug required");
+                }
+                return await callApi(apiBase, `/v1/projects/${encodeURIComponent(projectId)}`, {}, apiKey, { method: "PATCH", jsonBody: body });
+            }
+            case "delete_project": {
+                const projectId = safeArgs["projectId"];
+                if (typeof projectId !== "string" || projectId.length === 0 || projectId.length > 64) {
+                    return errorResponse("projectId required (1-64 chars)");
+                }
+                return await callApi(apiBase, `/v1/projects/${encodeURIComponent(projectId)}`, {}, apiKey, { method: "DELETE" });
+            }
+            case "get_account_health": {
+                // 2026-06-05 axis 4 Tier 1 = 自社 LLM infra 健康状態 サマリ。 既存
+                // 4 endpoint を 並列 fetch + 1 narrative 圧縮。 個別 fail は partial
+                // で carry (= 1 軸 timeout で summary を 止めない設計)。
+                const winRaw = typeof safeArgs["window"] === "string" ? safeArgs["window"] : "24h";
+                const window = winRaw === "1h" || winRaw === "24h" || winRaw === "7d" ? winRaw : "24h";
+                const now = Date.now();
+                const windowMs = window === "1h"
+                    ? 60 * 60 * 1000
+                    : window === "24h"
+                        ? 24 * 60 * 60 * 1000
+                        : 7 * 24 * 60 * 60 * 1000;
+                const startTime = new Date(now - windowMs).toISOString();
+                const endTime = new Date(now).toISOString();
+                const aggregateBody = { startTime, endTime, groupBy: "provider", metric: "count" };
+                const errorBody = { startTime, endTime, groupBy: "provider", metric: "error_rate" };
+                const costBody = { startTime, endTime, groupBy: "provider", metric: "cost" };
+                const percentileBody = { startTime, endTime, metric: "latency" };
+                const [countsRes, errorRes, costRes, percentilesRes, budgetRes, auditRes,] = await Promise.allSettled([
+                    callApi(apiBase, "/v1/query/aggregate", {}, apiKey, {
+                        method: "POST",
+                        jsonBody: aggregateBody,
+                    }),
+                    callApi(apiBase, "/v1/query/aggregate", {}, apiKey, {
+                        method: "POST",
+                        jsonBody: errorBody,
+                    }),
+                    callApi(apiBase, "/v1/query/aggregate", {}, apiKey, {
+                        method: "POST",
+                        jsonBody: costBody,
+                    }),
+                    callApi(apiBase, "/v1/query/percentiles", {}, apiKey, {
+                        method: "POST",
+                        jsonBody: percentileBody,
+                    }),
+                    callApi(apiBase, "/v1/account/llm-feature-budget", {}, apiKey),
+                    callApi(apiBase, "/v1/audit-log", { limit: 10 }, apiKey),
+                ]);
+                const extractJson = (r) => {
+                    if (r.status !== "fulfilled" || r.value.isError)
+                        return null;
+                    const txt = r.value.content[0]?.text ?? "";
+                    try {
+                        return JSON.parse(txt);
+                    }
+                    catch {
+                        return null;
+                    }
+                };
+                const counts = extractJson(countsRes);
+                const errors = extractJson(errorRes);
+                const cost = extractJson(costRes);
+                const percentiles = extractJson(percentilesRes);
+                const budget = extractJson(budgetRes);
+                const audit = extractJson(auditRes);
+                const totalCalls = counts?.total?.value ?? 0;
+                const errorRate = errors?.total?.value !== undefined && errors.total.value !== null
+                    ? errors.total.value
+                    : null;
+                const costUsd = cost?.total?.value ?? 0;
+                const p50 = percentiles?.p50 ?? null;
+                const p95 = percentiles?.p95 ?? null;
+                const p99 = percentiles?.p99 ?? null;
+                const budgetUsed = budget?.usedUsd ?? null;
+                const budgetLimit = budget?.monthlyLimitUsd ?? null;
+                const budgetPercent = budgetUsed !== null && budgetLimit !== null && budgetLimit > 0
+                    ? Math.round((budgetUsed / budgetLimit) * 1000) / 10
+                    : null;
+                const recentEvents = Array.isArray(audit?.events) ? audit.events.length : 0;
+                let summary = "ok";
+                if ((errorRate !== null && errorRate >= 0.1) ||
+                    (budgetPercent !== null && budgetPercent >= 90) ||
+                    (p95 !== null && p95 >= 10000)) {
+                    summary = "critical";
+                }
+                else if ((errorRate !== null && errorRate >= 0.03) ||
+                    (budgetPercent !== null && budgetPercent >= 70) ||
+                    (p95 !== null && p95 >= 3000)) {
+                    summary = "warn";
+                }
+                const partialFailures = [];
+                if (countsRes.status !== "fulfilled" || countsRes.value.isError)
+                    partialFailures.push("counts");
+                if (errorRes.status !== "fulfilled" || errorRes.value.isError)
+                    partialFailures.push("errorRate");
+                if (costRes.status !== "fulfilled" || costRes.value.isError)
+                    partialFailures.push("cost");
+                if (percentilesRes.status !== "fulfilled" || percentilesRes.value.isError)
+                    partialFailures.push("percentiles");
+                if (budgetRes.status !== "fulfilled" || budgetRes.value.isError)
+                    partialFailures.push("budget");
+                if (auditRes.status !== "fulfilled" || auditRes.value.isError)
+                    partialFailures.push("auditLog");
+                return {
+                    content: [
+                        {
+                            type: "text",
+                            text: JSON.stringify({
+                                window,
+                                totals: { calls: totalCalls, costUsd, errorRate },
+                                latency: { p50, p95, p99 },
+                                budget: { used: budgetUsed, limit: budgetLimit, percentUsed: budgetPercent },
+                                recentEvents,
+                                summary,
+                                partialFailures: partialFailures.length > 0 ? partialFailures : undefined,
+                            }),
+                        },
+                    ],
+                };
+            }
+            case "aggregate_calls": {
+                const body = {};
+                if (typeof safeArgs["startTime"] === "string")
+                    body["startTime"] = safeArgs["startTime"];
+                if (typeof safeArgs["endTime"] === "string")
+                    body["endTime"] = safeArgs["endTime"];
+                if (typeof safeArgs["groupBy"] === "string")
+                    body["groupBy"] = safeArgs["groupBy"];
+                if (typeof safeArgs["metric"] === "string")
+                    body["metric"] = safeArgs["metric"];
+                if (typeof safeArgs["provider"] === "string")
+                    body["provider"] = safeArgs["provider"];
+                if (typeof safeArgs["tagKey"] === "string")
+                    body["tagKey"] = safeArgs["tagKey"];
+                return await callApi(apiBase, "/v1/query/aggregate", {}, apiKey, {
+                    method: "POST",
+                    jsonBody: body,
+                });
+            }
+            case "list_audit_log": {
+                const q = {};
+                if (typeof safeArgs["limit"] === "number")
+                    q["limit"] = safeArgs["limit"];
+                if (typeof safeArgs["eventType"] === "string")
+                    q["eventType"] = safeArgs["eventType"];
+                if (typeof safeArgs["targetKind"] === "string")
+                    q["targetKind"] = safeArgs["targetKind"];
+                if (typeof safeArgs["actorUserId"] === "string")
+                    q["actorUserId"] = safeArgs["actorUserId"];
+                if (typeof safeArgs["from"] === "string")
+                    q["from"] = safeArgs["from"];
+                if (typeof safeArgs["to"] === "string")
+                    q["to"] = safeArgs["to"];
+                if (typeof safeArgs["cursor"] === "string")
+                    q["cursor"] = safeArgs["cursor"];
+                return await callApi(apiBase, "/v1/audit-log", q, apiKey);
+            }
+            case "list_saved_views": {
+                return await callApi(apiBase, "/v1/saved-views", {}, apiKey);
+            }
+            case "create_saved_view": {
+                const name = safeArgs["name"];
+                const filter = safeArgs["filter"];
+                if (typeof name !== "string" || name.length === 0 || name.length > 80) {
+                    return errorResponse("name required (1-80 chars)");
+                }
+                if (!filter || typeof filter !== "object") {
+                    return errorResponse("filter required (object)");
+                }
+                return await callApi(apiBase, "/v1/saved-views", {}, apiKey, {
+                    method: "POST",
+                    jsonBody: { name, filter },
+                });
+            }
+            case "delete_saved_view": {
+                const id = safeArgs["id"];
+                if (typeof id !== "string" || id.length === 0 || id.length > 64) {
+                    return errorResponse("id required (1-64 chars)");
+                }
+                return await callApi(apiBase, `/v1/saved-views/${encodeURIComponent(id)}`, {}, apiKey, { method: "DELETE" });
+            }
+            case "export_calls": {
+                const body = {};
+                if (typeof safeArgs["startTime"] === "string")
+                    body["startTime"] = safeArgs["startTime"];
+                if (typeof safeArgs["endTime"] === "string")
+                    body["endTime"] = safeArgs["endTime"];
+                if (typeof safeArgs["provider"] === "string")
+                    body["provider"] = safeArgs["provider"];
+                if (typeof safeArgs["model"] === "string")
+                    body["model"] = safeArgs["model"];
+                if (typeof safeArgs["limit"] === "number")
+                    body["limit"] = safeArgs["limit"];
+                return await callApi(apiBase, "/v1/query/export", {}, apiKey, {
+                    method: "POST",
+                    jsonBody: body,
+                });
+            }
+            case "bulk_delete_calls": {
+                const ids = safeArgs["callIds"];
+                if (!Array.isArray(ids) || ids.length === 0 || ids.length > 100) {
+                    return errorResponse("callIds required (non-empty string array, max 100)");
+                }
+                const strIds = [];
+                for (const id of ids) {
+                    if (typeof id !== "string" || id.length === 0 || id.length > 128) {
+                        return errorResponse("each callId must be non-empty string up to 128 chars");
+                    }
+                    strIds.push(id);
+                }
+                const body = { callIds: strIds };
+                if (safeArgs["dryRun"] === true)
+                    body["dryRun"] = true;
+                return await callApi(apiBase, "/v1/calls/bulk-delete", {}, apiKey, {
+                    method: "POST",
+                    jsonBody: body,
+                });
+            }
+            case "compare_eval_runs": {
+                const baselineId = validateAnnotationId(safeArgs["baselineRunId"]);
+                const candidateId = validateAnnotationId(safeArgs["candidateRunId"]);
+                if (!baselineId || !candidateId) {
+                    return errorResponse("baselineRunId + candidateRunId required (positive integers up to 10 digits)");
+                }
+                if (baselineId === candidateId) {
+                    return errorResponse("baselineRunId and candidateRunId must differ");
+                }
+                return await callApi(apiBase, "/v1/eval-runs/compare", { baseline: baselineId, candidate: candidateId }, apiKey);
+            }
             case "run_eval": {
                 const body = {};
                 if (typeof safeArgs["name"] === "string")