@aisa-one/cli 0.6.8 → 0.6.10

package/dist/commands/connect-terminal.js

@@ -1,6 +1,6 @@
 import * as readline from "node:readline/promises";
 import chalk from "chalk";
-import { AGENT_BADGE, AGENT_NOTES, BACKUP_COPY, CONFIRM, FILE_MODEL_FALLBACK, FILE_MODEL_NOTE, STEP_AGENT, STEP_CAPS, STEP_MODELS, STEP_TITLES, STEP_WELCOME, agentRank, fill, t, } from "./flow.js";
+import { AGENT_BADGE, AGENT_NOTES, BACKUP_COPY, CONFIRM, FILE_MODEL_FALLBACK, FILE_MODEL_NOTE, STEP_AGENT, STEP_CAPS, STEP_MODELS, STEP_TITLES, STEP_WELCOME, agentRank, t, } from "./flow.js";
 import { INSTALLERS } from "./install.js";
 import { defaultModelsFor } from "./llm-config.js";
 import { httpFetch } from "../utils/http.js";
@@ -612,104 +612,10 @@ export async function runTerminalFlow(o) {
                 // ── step 4: capabilities ──
                 if (from <= 4) {
                     console.log(header(4, o.lang));
-                    const totalTools = o.servers.reduce((n, s) => n + s.toolCount, 0);
-                    const cats = [...new Set(o.servers.map((s) => s.category))];
-                    say(fill(STEP_CAPS.counts, o.lang, {
-                        areas: cats.length, servers: o.servers.length, tools: totalTools,
-                    }));
-                    console.log(dim("│"));
-                    // Pull once more: the capability list is long, and a page open beside
-                    // this one is the likelier place to have ticked through it.
-                    const fresh = await pull(o);
-                    if (fresh?.draft?.servers?.length)
-                        draft.servers = fresh.draft.servers;
-                    if (fresh)
-                        rev = fresh.rev;
-                    const chosen = new Set(draft.servers);
-                    if (interactive()) {
-                        // Sorted by area, with the area named once where it changes.
-                        //
-                        // Twenty-five servers over eight areas — and five of those areas hold
-                        // exactly one server, so a heading per area produced "heading, row,
-                        // heading, row", which is a hat on every line rather than a grouping.
-                        // The name on the first row of each run says the same thing and costs
-                        // no rows: what follows inherits it from the blank space beside it.
-                        const ordered = [...o.servers].sort((a, b) => a.category.localeCompare(b.category) || a.slug.localeCompare(b.slug));
-                        const rowFor = (slug) => ordered.findIndex((s) => s.slug === slug);
-                        const initial = [...chosen].map(rowFor).filter((i) => i >= 0);
-                        const a3 = await pickOrWatch(o, rev, 4, ordered.map((srv) => ({
-                            label: srv.slug,
-                            meta: `${srv.toolCount} ${t(STEP_CAPS.toolsWord, o.lang)}`,
-                            tag: srv.category,
-                            // Straight from the host, the same sentence the page shows. A
-                            // hand-written table here would be a second source of truth for
-                            // something that ships with the server.
-                            detail: srv.description,
-                        })), true, initial, o.lang === "zh"
-                            ? "↑↓ 移动 · 空格勾选 · a 全选/全不选 · 回车确认 · esc 返回上一步"
-                            : "↑↓ move · space to tick · a for all · enter to confirm · esc to go back", 3, {
-                            read: (d) => d.servers === undefined ? undefined : d.servers.map(rowFor).filter((i) => i >= 0),
-                            write: async (indexes) => {
-                                const next = indexes.map((i) => ordered[i]?.slug).filter((x) => Boolean(x));
-                                ({ rev } = await push(o, rev, { draft: { servers: next } }));
-                                return rev;
-                            },
-                        });
-                        // In a checklist the way back cannot be another row — a row is a
-                        // thing you tick — so Escape is the gesture, and the hint says so.
-                        if (a3.by === "back")
-                            throw new Rewind(a3.to, a3.from === "here");
-                        if (a3.by === "page") {
-                            chosen.clear();
-                            for (const slug of a3.draft.servers ?? [])
-                                chosen.add(slug);
-                            const s2 = await pull(o);
-                            if (s2)
-                                rev = s2.rev;
-                            console.log("\n" + dim("│  ") + chalk.magenta(o.lang === "zh" ? "↩ 已在页面中选择" : "↩ chosen in the page"));
-                        }
-                        else {
-                            chosen.clear();
-                            for (const i of a3.picked ?? []) {
-                                const srv = ordered[i];
-                                if (srv)
-                                    chosen.add(srv.slug);
-                            }
-                        }
-                    }
-                    else {
-                        o.servers.forEach((s, i) => {
-                            const mark = chosen.has(s.slug) ? chalk.green("[x]") : dim("[ ]");
-                            console.log(`${dim("│")}  ${mark} ${bold(String(i + 1).padStart(2))}) ${s.slug.padEnd(24)} ${dim(String(s.toolCount) + " " + t(STEP_CAPS.toolsWord, o.lang))}`);
-                        });
-                        console.log(dim("│"));
-                        say(o.lang === "zh"
-                            ? "输入编号切换选中(空格分隔),直接回车确认。"
-                            : "Type numbers to toggle (space-separated), or press enter to confirm.");
-                        for (;;) {
-                            const answer = await askLine(bold.cyan("│  > "));
-                            if (answer === "")
-                                break;
-                            for (const tok of answer.split(/[\s,]+/)) {
-                                const n = Number(tok);
-                                if (!Number.isInteger(n) || n < 1 || n > o.servers.length)
-                                    continue;
-                                const slug = o.servers[n - 1].slug;
-                                if (chosen.has(slug))
-                                    chosen.delete(slug);
-                                else
-                                    chosen.add(slug);
-                            }
-                            console.log(dim("│  ") + chalk.green(`${chosen.size} selected: `) + dim([...chosen].join(", ")));
-                            ({ rev } = await push(o, rev, { draft: { servers: [...chosen] } }));
-                        }
-                    }
-                    draft.servers = [...chosen];
-                    if (draft.servers.length === 0) {
-                        console.log(dim("└─ ") + chalk.yellow(o.lang === "zh" ? "至少选一个" : "Pick at least one"));
-                        return undefined;
-                    }
-                    console.log(dim("└─ ") + chalk.green(`${draft.servers.length} ✓`));
+                    say(t(STEP_CAPS.ledeTail, o.lang));
+                    draft.servers = o.servers.map((s) => s.slug);
+                    for (const server of o.servers)
+                        console.log(dim("│  ") + server.name + " — " + server.endpoint);
                     // Publish the choice but NOT the step: arriving at step 5 is what makes
                     // the page start the run by itself, and the user has not confirmed yet.
                     // Announcing the step here applied everything without being asked.

package/dist/commands/connect.js

@@ -7,9 +7,9 @@ import { dirname, join } from "node:path";
 import chalk from "chalk";
 import { error } from "../utils/display.js";
 import { expandHome } from "../utils/file.js";
-import { MCP_CONFIGS, MCP_DEFAULT_SLUGS, AISA_PROVIDER_ID } from "../constants.js";
+import { MCP_CONFIGS, AISA_PROVIDER_ID } from "../constants.js";
 import { getAccessToken, getConfig, setConfig } from "../config.js";
-import { fetchLiveServers, writeClientConfig, buildEntry, stripped } from "./mcp.js";
+import { getMcpServers, writeClientConfig, buildEntry, stripped, isUnifiedMcpSelection } from "./mcp.js";
 import { INSTALLERS, installAgent, isInstalled, supported } from "./install.js";
 import { writeCodexLLM, writeClaudeCodeLLM, writeOpencodeLLM, writeCodexAisaProfile, writeOpencodeAisaBackup, defaultModelsFor, patchCodexMCPAuth, DEFAULT_MODELS, } from "./llm-config.js";
 import { writeClaudeAisaSettings, installWrappers } from "./wrappers.js";
@@ -35,6 +35,7 @@ import { handOverResults, closeStaleResults, SUPERSEDE_GRACE_MS } from "./serve-
 import { restoreTerminal, handOffStdin } from "./prompt.js";
 import { printBanner } from "../utils/banner.js";
 import { verifyKey } from "../utils/key-check.js";
+import { clientConfigKey } from "../utils/client-key.js";
 /**
  * Page templates. T1 is the original two-page flow (selection + live
  * progress, then a separate success page); T2 is the guided six-step flow.
@@ -62,12 +63,13 @@ export function resolveTemplate(flag) {
  *   sign in, exit. No daemon, no terminal takeover, no prompt or skill
  *   injection into the user's agent. The user stays in their own Claude Code.
  * - Sign-in is one OAuth round for everything: with no key stored, the run
- *   starts with the same browser approval `aisa login` uses. Every MCP entry
- *   is written with the current access token and the model provider gets
- *   the same token — zero
- *   per-server authorization popups. Only if that sign-in fails do we fall
- *   back to each client's own OAuth machinery (`claude mcp login <name>` per
- *   server; `codex mcp add` runs its own flow), which still works but costs
+ *   starts with the same browser approval `aisa login` uses, and the token it
+ *   yields is traded for the durable "aisa cli" key (POST /v1/keys/mint).
+ *   Every MCP entry is written with that key and the model provider gets
+ *   the same key — zero per-server authorization popups. Only if that
+ *   sign-in fails do we fall back to each client's own OAuth machinery
+ *   (`claude mcp login <name>` per server; `codex mcp add` runs its own
+ *   flow), which still works but costs
  *   one browser round per server. A key configured beforehand skips the
  *   sign-in entirely.
  * - The page reports the whole journey live (GET /status polling), and a
@@ -255,6 +257,8 @@ export function cursorDeeplink(name, endpoint, key) {
     return `cursor://anysphere.cursor-deeplink/mcp/install?name=${encodeURIComponent(name)}&config=${encodeURIComponent(config)}`;
 }
 async function applySelection(clientIds, chosen, key, dryRun, state) {
+    if (!isUnifiedMcpSelection(chosen))
+        throw new Error("Only the unified AIsa Router MCP can be configured");
     const results = [];
     for (const id of clientIds) {
         if (id === "cursor" && state?.deeplinks) {
@@ -375,7 +379,7 @@ async function applySelection(clientIds, chosen, key, dryRun, state) {
         }
         else if (MCP_CONFIGS[id]) {
             if (dryRun) {
-                results.push({ client: id, ok: true, message: `would write ${chosen.length + 1} entries to ${MCP_CONFIGS[id].path}` });
+                results.push({ client: id, ok: true, message: `would write ${chosen.length} entries to ${MCP_CONFIGS[id].path}` });
                 continue;
             }
             const r = writeClientConfig(id, chosen, key);
@@ -716,6 +720,26 @@ async function runPlan(state, input, log) {
     }
     // Reached the far side of the sign-in, so whatever it was is over.
     state.needsSignIn = false;
+    // Everything below writes into other programs' configs, which never refresh
+    // an OAuth token: trade it for the durable key first, or every entry dies
+    // with the token a day from now.
+    if (key && !input.dryRun) {
+        try {
+            key = await clientConfigKey(key);
+        }
+        catch (e) {
+            failures++;
+            const why = e.message.split("\n")[0].trim();
+            setStep(state, "mcp", { state: "fail", detail: `${why} — nothing was written` });
+            log.line("fail", "Issue a key for your clients", e.message);
+            for (const step of state.steps) {
+                if (step.state === "pending") {
+                    setStep(state, step.id, { state: "skip", detail: "waiting for a key" });
+                }
+            }
+            return failures;
+        }
+    }
     // ── MCP entries ──
     const willAuthorize = !key && input.clients[0] === "codex";
     setStep(state, "mcp", {
@@ -1109,14 +1133,13 @@ function renderPage(servers, clients, token, keyed, canInstall) {
         .map(([cat, list]) => {
         const rows = list
             .map((s) => {
-            const checked = MCP_DEFAULT_SLUGS.includes(s.slug) ? "checked" : "";
             // The full manifest description, CSS-clamped to two lines; "more"
             // expands it and is hidden by the page script when the text already
             // fits — so short cards carry no dead link and long ones stay tidy.
-            return `<label class="card ${checked ? "on" : ""}" data-kind="server">
-  <input type="checkbox" name="server" value="${s.slug}" ${checked}>
+            return `<label class="card on" data-kind="server">
+  <input type="checkbox" name="server" value="${s.slug}" checked disabled>
   <span class="body"><span class="head"><span class="name">${stripped(s.name)}</span>
-    <span class="badge">${s.toolCount} tools</span></span>
+    </span>
     <span class="desc">${s.description}</span>
     <span class="more" data-more>more</span></span></label>`;
         })
@@ -1147,24 +1170,21 @@ function renderPage(servers, clients, token, keyed, canInstall) {
         (rest.length
             ? `<div class="chips">${rest.map((c) => `${c.label} <i>· not found</i>`).join(" &nbsp;&nbsp; ")}</div>`
             : "");
-    const totalTools = servers.reduce((n, s) => n + s.toolCount, 0);
     // Named per client so the page can say which model a pick actually gets.
     const modelByClient = Object.fromEntries(clients.map((c) => [c.id, defaultModelsFor(c.id).model]));
     const authCopy = keyed
         ? `Your current AIsa credential is written into each entry — <b>no sign-in needed</b>.`
         : `<b>One sign-in, nothing to paste.</b> Your browser opens the AIsa approval
-       <b>once</b>; every server and model below is configured with the current
-       access token. Reconnect when that token expires.`;
+       <b>once</b>; it issues a long-lived key for this machine, and every server
+       and model below is configured with it — no further popups.`;
     const body = `
 <div class="eyebrow">Connect</div>
 <h1><em>AIsa MCP</em> — powerful real-world reach for your agent</h1>
-<p class="lede">One connection puts <b>${totalTools} live tools</b> — web search &amp; research,
-X/Twitter, Reddit, Instagram, stocks, crypto, prediction markets and B2B data — inside the
-coding agent you already use. Pick what you need, press Connect, approve in the browser. Done.</p>
+<p class="lede">One connection gives your coding agent access to AIsa tools. Press Connect and approve in the browser. Done.</p>
 
 <div class="cols">
 <div class="left">
-<h2><span class="n">1</span>Choose capabilities</h2>
+<h2><span class="n">1</span>Connect AIsa Router</h2>
 ${serverGroups}
 </div>
 
@@ -1439,7 +1459,7 @@ machine except the OAuth you approve. The process exits when everything is conne
     var servers = picked("server");
     var chosen = document.querySelector('input[name="client"]:checked');
     if (!servers.length || !chosen) {
-      result.textContent = "Pick at least one capability and one client."; return;
+      result.textContent = "Choose a client to connect to AIsa Router."; return;
     }
     var clients = [chosen.value];
     var install = chosen.dataset.install === "1" ? [chosen.value] : [];
@@ -1502,9 +1522,6 @@ export function renderDone(chosen, clientIds, steps, allServers, balanceMicros,
        Fix the above, then run <code>npx @aisa-one/cli connect</code> again — it is
        safe to re-run and picks up where things stand.</div></div>`
         : "";
-    const toolCount = chosen.reduce((n, s) => n + s.toolCount, 0);
-    const remaining = allServers.length - chosen.length;
-    const remainingTools = allServers.reduce((n, s) => n + s.toolCount, 0) - toolCount;
     // A run that installed an agent is a much bigger deal than one that only
     // added tools, and the headline must say so: the user walked in with
     // nothing and walked out with a working coding agent on AIsa models.
@@ -1515,12 +1532,12 @@ export function renderDone(chosen, clientIds, steps, allServers, balanceMicros,
     const model = defaultModelsFor(clientIds[0] ?? "");
     const successHeadline = installedAgents.length
         ? `<h1>Congratulations — <em>${installedAgents.join(" & ")}</em> is installed and armed
-with <em>${toolCount} powerful tool${toolCount > 1 ? "s" : ""}</em></h1>
+with <em>AIsa tools</em></h1>
 <p class="lede"><b>${installedAgents.join(" & ")}</b> is now on this machine, signed in to
 AIsa${llmOk ? `, running on <b>${model.model}</b> through AIsa,` : ","} with
 ${chosen.length} MCP server${chosen.length > 1 ? "s" : ""} wired in — a complete AI setup,
 nothing else to configure.</p>`
-        : `<h1>Congratulations — your agent just got <em>${toolCount} powerful new tool${toolCount > 1 ? "s" : ""}</em></h1>
+        : `<h1>Congratulations — your agent just got <em>access to AIsa tools</em></h1>
 <p class="lede">${chosen.length} AIsa MCP server${chosen.length > 1 ? "s are" : " is"} now
 installed and signed in for <b>${clientNames}</b> — nothing else to configure.</p>`;
     // Launchable only for the CLI agents, and only while the local process is
@@ -1596,7 +1613,7 @@ ${successHeadline}`;
     const recap = `
 <h2>Everything you just gained</h2>
 <div class="recap">
-  <div class="rsum">${chosen.length} capabilit${chosen.length === 1 ? "y" : "ies"} · ${toolCount} tools · ${clientNames}</div>
+  <div class="rsum">AIsa Router MCP · ${clientNames}</div>
 ${recapRows}
 </div>`;
     // ── the balance card: always shown, always with the top-up button — money
@@ -1674,10 +1691,7 @@ ${headline}
 ${mcpFailed
         ? failBlock + recap + balanceCard
         : `<p class="lede" style="margin-top:.6rem">You are now connected to <b>AIsa</b> — a powerful capability layer for agents: one account for
-<b>Claude, GPT, Gemini, DeepSeek, Kimi, GLM</b> and the live data behind them: <b>Twitter, SimilarWeb, CoinGecko, Apollo, Polymarket</b> and more.${remaining > 0
-            ? ` ${remaining} more MCP server${remaining > 1 ? "s" : ""} (${remainingTools} tools) are one
-<code>npx @aisa-one/cli connect</code> away.`
-            : ""}
+<b>Claude, GPT, Gemini, DeepSeek, Kimi, GLM</b> and the live data behind them: <b>Twitter, SimilarWeb, CoinGecko, Apollo, Polymarket</b> and more.
 Explore the platform at <a href="https://aisa.one" target="_blank" rel="noopener noreferrer">aisa.one</a> ·
 usage &amp; billing at <a href="https://console.aisa.one" target="_blank" rel="noopener noreferrer">console.aisa.one</a>.</p>
 ${failBlock}
@@ -2224,14 +2238,17 @@ export async function connectAction(options) {
     }
     let servers;
     try {
-        servers = resumed ? resumed.servers : await fetchLiveServers();
+        servers = resumed ? resumed.servers : getMcpServers();
+        if (!isUnifiedMcpSelection(servers))
+            throw new Error("This saved run uses retired MCP endpoints. Start again with aisa connect --force.");
     }
     catch (e) {
-        log.line("fail", "Could not read the MCP manifest", e.message);
-        log.note("check your network and try again");
+        log.line("fail", "Could not prepare the unified MCP connection", e.message);
+        log.note("Review your clients and models again with: aisa connect --force");
         process.exitCode = 1;
         return;
     }
+    log.note("Existing MCP entries are preserved. Remove old domain entries manually after connecting to aisa-router.");
     const clients = resumed ? resumed.clients : detectClients();
     // Rehearsal aid: in a dry run, pretend the listed clients are present so a
     // page can be reviewed for a client this machine does not have. Never
@@ -2291,7 +2308,7 @@ export async function connectAction(options) {
         // that never opens a browser starts from the same defaults rather than
         // from nothing.
         draft: {
-            servers: servers.filter((s) => MCP_DEFAULT_SLUGS.includes(s.slug)).map((s) => s.slug),
+            servers: servers.map((s) => s.slug),
             clients: [],
             install: [],
             llmMode: "backup",
@@ -2534,7 +2551,16 @@ export async function connectAction(options) {
                 res.writeHead(403).end();
                 return;
             }
-            res.writeHead(200, { "content-type": "application/json" }).end(JSON.stringify({ key: key ?? null }));
+            // Read now, not at startup: the run may have signed in since. And the
+            // durable key, never the OAuth token — it is about to be pasted into
+            // an app that will not refresh it.
+            const current = await getAccessToken();
+            let pasted = null;
+            try {
+                pasted = current ? await clientConfigKey(current) : null;
+            }
+            catch { /* no key to offer; the page says so */ }
+            res.writeHead(200, { "content-type": "application/json" }).end(JSON.stringify({ key: pasted }));
             return;
         }
         if (req.method === "POST" && url.pathname === "/lang") {
@@ -2684,7 +2710,7 @@ export async function connectAction(options) {
             }
             // Who started this decides whether there is an animation to wait for.
             drivenByTerminal = req.headers["x-connect-source"] === "terminal";
-            chosenServers = servers.filter((s) => body.servers?.includes(s.slug));
+            chosenServers = servers;
             chosenClients = body.clients ?? [];
             const wantInstall = new Set(body.install ?? []);
             // Ticking "install" means "and connect it": nobody installs an agent
@@ -2964,7 +2990,7 @@ export async function connectAction(options) {
     // the real problem.
     if (resumed && state.selection) {
         chosenClients = [...state.selection.clients];
-        chosenServers = servers.filter((x) => state.selection.servers.includes(x.slug));
+        chosenServers = servers;
         // llmMode is not a local here — the routes read state.llmMode, which
         // travels with the run.
     }
@@ -2975,7 +3001,7 @@ export async function connectAction(options) {
         armRetry({
             install: [...sel.install],
             clients: [...sel.clients],
-            servers: servers.filter((x) => sel.servers.includes(x.slug)),
+            servers: servers,
             key: liveKey(),
             catcher: signInCatcher,
             signinShown: signinShownP,
@@ -3095,7 +3121,7 @@ export async function connectAction(options) {
     const missing = clients.filter((c) => !c.detected).map((c) => c.label);
     if (missing.length)
         log.record(`not found: ${missing.join(", ")}`);
-    log.line("info", `${servers.length} AIsa MCP servers are live`, "pick what you need");
+    log.line("info", "AIsa Router MCP", "one connection for all published tools");
     // The terminal walks the same six steps from the same definition. It is not
     // a fallback: on a machine with no browser, printing a localhost URL is a
     // dead end, and both surfaces share one draft so either can be used —

package/dist/commands/flow.d.ts

@@ -262,51 +262,13 @@ export declare const STEP_CAPS: {
         en: string;
         zh: string;
     };
-    pickArea: {
-        en: string;
-        zh: string;
-    };
-    selectAll: {
-        en: string;
-        zh: string;
-    };
-    /**
-     * The counts sentence as one string, because punctuation is not shared
-     * between languages: joining "N", "areas", ", " in code produced
-     * "2 个领域, 2 个 server, 48 个工具." — Chinese text with ASCII commas and a
-     * full stop. Whatever varies goes in as a placeholder instead.
-     */
     counts: {
         en: string;
         zh: string;
     };
-    /** Category tile meta: "3 servers · 47 tools". */
-    tileMeta: {
-        en: string;
-        zh: string;
-    };
-    /**
-     * A placeholder both renderers can fill. The page used to carry
-     * <b id="mModel"></b> here and fill it with script — which left the
-     * terminal printing "AIsa()" with nothing inside, because stripping tags is
-     * all it can do with markup meant for the other surface.
-     */
-    serversWord: {
-        en: string;
-        zh: string;
-    };
-    serverWord: {
-        en: string;
-        zh: string;
-    };
-    toolsWord: {
-        en: string;
-        zh: string;
-    };
 };
 /** Shown for a file-configured client that has no note of its own. */
 export declare const FILE_MODEL_FALLBACK: Text;
-export declare const CATEGORY_BLURB: Record<string, Text>;
 export declare const STEP_INSTALL: {
     h1: {
         en: string;
@@ -340,10 +302,6 @@ export declare const STEP_DONE: {
     };
 };
 export declare const RUNTIME_COPY: {
-    readonly nothingSelected: {
-        readonly en: "Nothing selected yet — pick at least one server.";
-        readonly zh: "还没有选择 —— 至少选一个 server。";
-    };
     readonly planBackup: {
         readonly "claude-code": {
             readonly en: "Install the claude-aisa command";

package/dist/commands/flow.js

@@ -136,12 +136,12 @@ export const AGENT_NOTES = {
         zh: "每个 server 一次点击:最后一步会给出 <b>Add to Cursor</b> 链接。Cursor 会打开、把确切的配置内容给你看,你确认后由它自己写进 MCP 设置。模型保持原样 —— Cursor 用它自己的。",
     },
     "claude-desktop": {
-        en: "Claude Desktop's config file only takes MCP servers over stdio, so each AIsa server is written in as a small <code>mcp-remote</code> bridge carrying your key (plus the free <code>aisa-docs</code> server). The bridges live and die with the app — no background service, no open ports. Models cannot be changed: Claude Desktop runs Anthropic's own. Restart the app and the servers appear under its tools menu.",
-        zh: "Claude Desktop 的配置文件只接受 stdio 形式的 MCP server,所以每个 AIsa server 都写成一个带你 key 的 <code>mcp-remote</code> 小桥接(外加免费的 <code>aisa-docs</code>)。桥接随 app 启停 —— 没有后台服务,不开端口。模型无法更改:Claude Desktop 跑的是 Anthropic 自家模型。重启 app 后就能在工具菜单里看到这些 server。",
+        en: "Claude Desktop's config file only takes MCP servers over stdio, so the AIsa Router is written in as a small <code>mcp-remote</code> bridge carrying your key. The bridge lives and dies with the app — no background service, no open ports. Models cannot be changed: Claude Desktop runs Anthropic's own. Restart the app and the Router appears under its tools menu.",
+        zh: "Claude Desktop 的配置文件只接受 stdio 形式的 MCP server,所以 AIsa Router 写成一个带你 key 的 <code>mcp-remote</code> 小桥接。桥接随 app 启停 —— 没有后台服务,不开端口。模型无法更改:Claude Desktop 跑的是 Anthropic 自家模型。重启 app 后就能在工具菜单里看到 AIsa Router。",
     },
     "claude-ai": {
-        en: "Nothing is installed. claude.ai takes remote MCP servers as <b>Connectors</b>: on the last step you get one URL per server with a Copy button, and add each under <b>Settings → Connectors → Add custom connector</b>, then press Connect — claude.ai runs the AIsa sign-in itself. Models cannot be changed: claude.ai runs Anthropic's own.",
-        zh: "不安装任何东西。claude.ai 以 <b>Connectors</b> 的形式接受远程 MCP server:最后一步会给出每个 server 的 URL 和复制按钮,在 <b>Settings → Connectors → Add custom connector</b> 里逐个添加,然后点 Connect —— claude.ai 会自己跑 AIsa 的登录流程。模型无法更改:claude.ai 跑的是 Anthropic 自家模型。",
+        en: "Nothing is installed. claude.ai takes remote MCP servers as <b>Connectors</b>: on the last step you get the Router URL with a Copy button, and add it under <b>Settings → Connectors → Add custom connector</b>, then press Connect — claude.ai runs the AIsa sign-in itself. Models cannot be changed: claude.ai runs Anthropic's own.",
+        zh: "不安装任何东西。claude.ai 以 <b>Connectors</b> 的形式接受远程 MCP server:最后一步会给出 Router 的 URL 和复制按钮,在 <b>Settings → Connectors → Add custom connector</b> 里添加,然后点 Connect —— claude.ai 会自己跑 AIsa 的登录流程。模型无法更改:claude.ai 跑的是 Anthropic 自家模型。",
     },
 };
 /**
@@ -176,8 +176,8 @@ export const AGENT_HAVE_NOTES = {
         zh: "你原有的东西一样都不会被替换。那两个文件里其他的 MCP server 和模型分组保持原样;只有 <code>aisa-*</code> 条目和 AIsa 那个分组是我们写的,要移除就是删掉它们。",
     },
     cursor: {
-        en: "Nothing of yours is replaced. Cursor adds each AIsa server beside your existing MCP entries under <code>aisa-*</code> names, and you approve every one inside Cursor first.",
-        zh: "你原有的东西一样都不会被替换。Cursor 会把每个 AIsa server 以 <code>aisa-*</code> 命名加在你已有的 MCP 条目旁边,而且每一个都要你在 Cursor 里先确认。",
+        en: "Nothing of yours is replaced. Cursor adds the AIsa Router beside your existing MCP entries under the <code>aisa-router</code> name, and you approve it inside Cursor first.",
+        zh: "你原有的东西一样都不会被替换。Cursor 会把 AIsa Router 以 <code>aisa-router</code> 命名加在你已有的 MCP 条目旁边,并需要你在 Cursor 里先确认。",
     },
 };
 // ── the rail: every step's name ─────────────────────────────────────────────
@@ -325,81 +325,21 @@ export const BACKUP_COPY = {
 };
 // ── step 4: capabilities ────────────────────────────────────────────────────
 export const STEP_CAPS = {
-    h1: { en: "What should your agent be able to reach?", zh: "你的 agent 需要够得着什么?" },
+    h1: { en: "Connect AIsa tools", zh: "连接 AIsa 工具" },
     ledeTail: {
-        en: "Open an area to pick the servers inside it. Everything here is live production data with licensed sources; you can add more later with one more <code>aisa connect</code>.",
-        zh: "点开一个领域来挑里面的 server。这里全都是有授权来源的生产环境实时数据;之后再跑一次 <code>aisa connect</code> 就能加更多。",
+        en: "One Router MCP connects your agent to AIsa tools. Existing MCP entries stay unchanged; remove old domain entries manually.",
+        zh: "通过一个 Router MCP 接入 AIsa 工具。已有 MCP 条目保持不变；旧的领域条目请自行删除。",
     },
-    pickArea: { en: "Pick an area above", zh: "先在上面选一个领域" },
-    selectAll: { en: "Select all in this area", zh: "全选这个领域" },
-    /**
-     * The counts sentence as one string, because punctuation is not shared
-     * between languages: joining "N", "areas", ", " in code produced
-     * "2 个领域, 2 个 server, 48 个工具." — Chinese text with ASCII commas and a
-     * full stop. Whatever varies goes in as a placeholder instead.
-     */
     counts: {
-        en: "{areas} areas, {servers} servers, {tools} tools.",
-        zh: "{areas} 个领域、{servers} 个 server、{tools} 个工具。",
+        en: "One unified AIsa MCP connection.",
+        zh: "一个统一的 AIsa MCP 连接。",
     },
-    /** Category tile meta: "3 servers · 47 tools". */
-    tileMeta: {
-        en: "{n} {noun} · {tools} tools",
-        zh: "{n} 个 {noun} · {tools} 个工具",
-    },
-    /**
-     * A placeholder both renderers can fill. The page used to carry
-     * <b id="mModel"></b> here and fill it with script — which left the
-     * terminal printing "AIsa()" with nothing inside, because stripping tags is
-     * all it can do with markup meant for the other surface.
-     */
-    serversWord: { en: "servers", zh: "server" },
-    serverWord: { en: "server", zh: "server" },
-    toolsWord: { en: "tools", zh: "个工具" },
 };
 /** Shown for a file-configured client that has no note of its own. */
 export const FILE_MODEL_FALLBACK = {
     en: "This client picks its own models inside the app, so nothing is changed here — only the MCP servers are added.",
     zh: "这个客户端在它自己的应用里选模型,所以这一步不做改动 —— 只添加 MCP server。",
 };
-export const CATEGORY_BLURB = {
-    "Search & Research": {
-        en: "Ranked web results with page text already extracted, plus YouTube.",
-        zh: "带排序的网页搜索结果,正文已经抽好,另有 YouTube。",
-    },
-    Finance: {
-        en: "US equities, crypto, prediction markets and what X is saying about a ticker.",
-        zh: "美股、加密货币、预测市场,以及 X 上关于某只票的讨论。",
-    },
-    Social: {
-        en: "Public X/Twitter, Reddit, Instagram and Pinterest — profiles, posts, engagement.",
-        zh: "公开的 X/Twitter、Reddit、Instagram 和 Pinterest —— 资料、帖子、互动数据。",
-    },
-    Sales: {
-        en: "Apollo B2B data — enrich people and companies, find prospects.",
-        zh: "Apollo 的 B2B 数据 —— 补全人和公司信息、找潜在客户。",
-    },
-    "SEO & Search Data": {
-        en: "Keywords, rankings, backlinks and on-page checks — and how a brand shows up in AI answers.",
-        zh: "关键词、排名、外链、页面检查 —— 还有一个品牌在 AI 回答里的露出情况。",
-    },
-    Communication: {
-        en: "An inbox an agent can actually use — its own address, sending and receiving.",
-        zh: "agent 能真正用起来的邮箱 —— 自己的地址,能收能发。",
-    },
-    Creators: {
-        en: "Find creators by audience, topic and reach, across platforms.",
-        zh: "按受众、话题和影响力找创作者,跨平台。",
-    },
-    Search: {
-        en: "YouTube search on its own.",
-        zh: "单独的 YouTube 搜索。",
-    },
-    "Market Intelligence": {
-        en: "Similarweb data — a site's traffic, engagement and where its visitors come from.",
-        zh: "Similarweb 数据 —— 网站流量、互动情况,以及访客从哪来。",
-    },
-};
 // ── step 5: install ─────────────────────────────────────────────────────────
 export const STEP_INSTALL = {
     h1: { en: "Ready to connect", zh: "准备接线" },
@@ -412,8 +352,8 @@ export const STEP_INSTALL = {
         zh: "你当前的 AIsa 凭证会写进每一条配置 —— <b>无需登录</b>。",
     },
     authFresh: {
-        en: "<b>One sign-in, nothing to paste.</b> Your browser opens the AIsa approval once; every server and model is configured with the current access token. Reconnect when that token expires.",
-        zh: "<b>登录一次,不用粘贴任何东西。</b>浏览器会打开一次 AIsa 授权页;所有 server 和模型都使用当前 access token 配置。token 过期后请重新连接。",
+        en: "<b>One sign-in, nothing to paste.</b> Your browser opens the AIsa approval once; it issues a long-lived key for this machine, and every server and model is configured with it — no further popups.",
+        zh: "<b>登录一次,不用粘贴任何东西。</b>浏览器会打开一次 AIsa 授权页;它会为这台机器签发一把长期有效的 key,所有 server 和模型都用它配置 —— 不会再有弹窗。",
     },
 };
 // ── step 6: done ────────────────────────────────────────────────────────────
@@ -426,10 +366,6 @@ export const STEP_DONE = {
 // Resolved to the current language and injected as a single object, because
 // the browser has no access to this file. Same source, one hop later.
 export const RUNTIME_COPY = {
-    nothingSelected: {
-        en: "Nothing selected yet — pick at least one server.",
-        zh: "还没有选择 —— 至少选一个 server。",
-    },
     planBackup: {
         "claude-code": { en: "Install the claude-aisa command", zh: "安装 claude-aisa 命令" },
         codex: { en: "Add the aisa profile and codex-aisa", zh: "添加 aisa profile 和 codex-aisa" },