@mediamonks/canopy-cli 0.7.29 → 0.7.31

package/dist/docs/cli-mcp.md

@@ -0,0 +1,88 @@
+---
+title: canopy mcp
+description: Let AI agents use a project's backend as MCP tools — check access, print client config, and run the stdio server.
+---
+
+`canopy mcp` connects AI agents (Claude Code, Claude Desktop, Cursor, any MCP
+client) to a project's tools. The agent always acts as you. See
+[Agent tools](https://docs.canopy.website/guides/agent-tools) for how a project offers tools.
+
+## Usage
+
+```bash
+canopy mcp add [slug] [--channel staging]
+canopy mcp serve [slug] [--channel staging] [--local]
+```
+
+## canopy mcp add
+
+Checks that you're signed in and may open the project, reports how many tools
+it serves, and prints configuration for your client: a `claude mcp add` line
+for Claude Code, and an `mcpServers` JSON block for Claude Desktop, Cursor and
+others. It writes nothing; copy what you need.
+
+The printed command names the site (`--profile`) and, for staging, the
+channel, so it keeps working whatever your default site is. It contains no
+credentials.
+
+## canopy mcp serve
+
+The MCP server your client starts. It talks MCP on stdin/stdout and writes
+nothing else there, so everything it has to say goes to stderr, where clients
+show it as the server's log.
+
+It uses your `canopy login`: it exchanges it for short-lived credentials for
+this one project, keeps them in memory, and renews them as needed. If you're
+not signed in, it exits with a message telling you to run `canopy login`.
+
+Only the project's tools reach the agent, and the tool list stays current: a
+redeploy with new tools reaches a running session without restarting
+anything.
+
+## Arguments & flags
+
+- `slug` — the project. Defaults to the `canopy.json` in the current directory.
+- `--channel <live|staging>` — which channel's backend to use. Default `live`.
+- `--local` (serve only) — use the tools of the `canopy dev` running in the
+  current directory instead of the deployed project.
+- `--profile <id>` — which Canopy site, as for every command.
+
+## Examples
+
+```bash
+canopy mcp add notes
+```
+
+Check you can use `notes` and print its client configuration.
+
+```bash
+claude mcp add canopy-notes -- canopy mcp serve notes --profile main
+```
+
+Add `notes` to Claude Code (the line `canopy mcp add` printed).
+
+```bash
+claude mcp add notes-staging -- canopy mcp serve notes --channel staging --profile main
+```
+
+Try the staging channel's tools before promoting them.
+
+```bash
+claude mcp add notes-local -- canopy mcp serve --local
+```
+
+While developing: the tools from your working tree, served by `canopy dev`.
+
+## When it can't connect
+
+`canopy mcp add` and `canopy mcp serve` say what's wrong in plain words. The
+usual cases:
+
+- **Not signed in** — run `canopy login`.
+- **Limited to its collaborators** — ask the owner to add you.
+- **No backend, or serves no tools** — the project has no `worker/`, or it
+  doesn't mount any tools yet.
+- **Redeploy required** — the backend was deployed before tools existed. Run
+  `canopy deploy`.
+- **Invalid tool definitions** — two tools share a name, or a schema is wrong.
+  The message names the tool.

package/dist/docs/cli.md

@@ -60,6 +60,8 @@ has a full page per command.
   `--snapshot <key>`. `--yes` skips the confirmation. **Destructive.**
 - `canopy usage` — AI spend and storage for the project (same data as the
   dashboard Usage tab). `--project <slug>` / `--json` for agents.
+- `canopy ai models` — the AI models and `auto/*` aliases this site offers,
+  with prices. Works from any directory; `--json` for agents.
 
 ## CI / deploy tokens
 

package/dist/docs/developers-agents-and-ai.md

@@ -64,6 +64,14 @@ already knows the rules:
   use `canopy dev --local` for an offline sandbox where writes never reach
   the platform.
 
+## Give agents your project's tools
+
+The other way round: your project can offer its backend to agents as MCP
+tools, so a colleague's Claude Code or Cursor can call it (as them) without
+opening the app. Wrap the routes you already have with `mountMcpTools`, deploy,
+and people connect with `canopy mcp add <slug>`. See
+[Agent tools](https://docs.canopy.website/guides/agent-tools).
+
 ## Prompts that work
 
 With the skill installed, these are safe, copy-pasteable asks — each maps to

package/dist/docs/index.json

@@ -67,6 +67,11 @@
           "title": "CI and deploy tokens",
           "summary": "Push-to-main deploys using a project-scoped deploy token."
         },
+        {
+          "slug": "agent-tools",
+          "title": "Agent tools (MCP)",
+          "summary": "Let AI agents like Claude Code and Cursor use your project's backend as tools — as the person running them."
+        },
         {
           "slug": "porting-nextjs",
           "title": "Porting a Next.js app",
@@ -276,11 +281,21 @@
           "title": "canopy usage",
           "summary": "Show a project's AI spend, storage footprint, and deploy count from the terminal."
         },
+        {
+          "slug": "cli-ai",
+          "title": "canopy ai models",
+          "summary": "List the AI models and auto/* aliases your site offers, with prices."
+        },
         {
           "slug": "cli-feedback",
           "title": "canopy feedback",
           "summary": "Read and triage a project's feedback inbox, or file a bug/feedback report — against a project or against Canopy itself."
         },
+        {
+          "slug": "cli-mcp",
+          "title": "canopy mcp",
+          "summary": "Let AI agents use a project's backend as MCP tools — check access, print client config, and run the stdio server."
+        },
         {
           "slug": "cli-open",
           "title": "canopy open",

package/dist/docs/platform-ai-usage.md

@@ -21,10 +21,12 @@ const reply = await canopy.ai.chat({
 
 - **No keys in your code.** Requests authenticate as the project; the
   platform holds the provider credentials.
-- **A curated model list.** Each site enables a set of models
-  (`provider/model`, e.g. `anthropic/claude-haiku-4-5`) with a sensible
-  default — ask for a model outside the list and the API tells you what _is_
-  available.
+- **Pick a tier, not a model.** Name `auto/low` (the default), `auto/mid`, or
+  `auto/high` and the platform answers with the best current model for it.
+  You can also name a specific model from the site's list. `canopy ai models`
+  and the Models section on your project's Usage tab show it, with prices.
+  Names that aren't on the list are refused, and retired models keep working,
+  answered by their replacement, until you switch.
 - **Streaming** via `canopy.ai.chatStream()` for chat-style UIs.
 - **Documents and images.** Messages can attach files — reference something
   in the project's [blob storage](https://docs.canopy.website/sdk/blob) by key, or inline base64 — see

package/dist/docs/sdk-ai.md

@@ -8,6 +8,29 @@ Every request authenticates as the project and is metered against hard
 per-project budgets — spend, limits, and how prompts are handled by the
 model providers are covered in [AI access & usage](https://docs.canopy.website/platform/ai-usage).
 
+## Picking a model
+
+Name a quality tier instead of a model id, and the platform answers with the
+best current model for it:
+
+- `auto/low` — fast and cheap. The **default** when you leave `model` out.
+- `auto/mid` — a good balance for most product features.
+- `auto/high` — the most capable model, for hard reasoning and long documents.
+
+Each alias moves to newer models as they arrive, without a code change. You
+can also name a specific model from the site's list. Run
+`canopy ai models` to see the aliases, what each answers with right now, and
+every model with its price:
+
+```bash
+canopy ai models
+```
+
+Only names on that list work: anything else is refused with a `400` that lists
+what is available. When a model is retired, calls naming it keep working,
+answered by its replacement: `res.model` names the model that answered, and
+the response carries a `deprecation` notice (below) until you switch.
+
 ## ai.chat(options)
 
 One-shot chat completion. Takes a `ChatOptions` object, resolves to a
@@ -19,7 +42,7 @@ const res = await canopy.ai.chat({
     { role: "system", content: "You are terse." },
     { role: "user", content: "Say hi in five words." },
   ],
-  model: "anthropic/claude-haiku-4-5", // optional
+  model: "auto/mid", // optional — defaults to auto/low
   temperature: 0.7, // optional, 0–2
   maxTokens: 256, // optional
 });
@@ -34,10 +57,10 @@ res.usage; // { inputTokens: number, outputTokens: number }
 - `messages` — required, non-empty array of `{ role, content }`. `role` is
   `"system" | "user" | "assistant"`; `content` is a non-empty string, or an
   array of content parts to attach documents and images (below).
-- `model` — optional `"{provider}/{model}"` id. Omit it and the platform's
-  default model answers. Ask for a model whose provider isn't enabled and the
-  call fails with a message listing what _is_ available; `ai.endpoint()`
-  returns the full curated list.
+- `model` — optional: an alias (`auto/low`, `auto/mid`, `auto/high`) or a
+  `"{provider}/{model}"` id from the site's list (`canopy ai models`). Omit it
+  and `auto/low` answers. A name that isn't on the list fails with a message
+  listing what _is_ available.
 - `temperature` — optional number between 0 and 2. Reasoning models (the
   GPT-5 family) only accept their default temperature, so the platform drops
   this setting for them rather than failing the call — it has effect on
@@ -52,7 +75,11 @@ res.usage; // { inputTokens: number, outputTokens: number }
 
 **`ChatResponse`:** `{ message, model, usage: { inputTokens, outputTokens } }`.
 `message.role` is `"assistant"` and `message.content` is always a plain
-string, whatever went in.
+string, whatever went in. `model` is the model that actually answered. When
+you named a retired model, the response also has
+`deprecation: { requested, replacement }`. Switch to the replacement or an
+alias. Streams and the endpoints below say the same with a `Deprecation:
+true` response header.
 
 ## Attaching documents and images
 
@@ -143,8 +170,10 @@ The compat endpoint speaks OpenAI's chat-completions dialect to every model —
 portable, but the lowest common denominator. When you want a provider's own
 protocol — Anthropic prompt caching, PDFs with citations, or OpenAI's
 Responses API — use the native base URLs from `endpoint()`. Requests pass
-through to the provider unchanged; auth, the model allowlist, the daily
-budget, and metering all still apply.
+through to the provider unchanged; auth, the model list, the daily budget,
+and metering all still apply. Aliases work here too: on the Anthropic endpoint
+`auto/mid` picks the best Claude model for that tier, on the OpenAI endpoint
+the best OpenAI one.
 
 ```ts
 import Anthropic from "@anthropic-ai/sdk";
@@ -152,7 +181,7 @@ import Anthropic from "@anthropic-ai/sdk";
 const { native } = await canopy.ai.endpoint();
 const anthropic = new Anthropic({ baseURL: native.anthropic, apiKey: "unused" });
 const msg = await anthropic.messages.create({
-  model: "claude-haiku-4-5", // bare or "anthropic/…" — must be on the curated list
+  model: "auto/mid", // an alias, or a model id from `canopy ai models` (bare or "anthropic/…")
   max_tokens: 1024,
   messages: [{ role: "user", content: "Hello" }],
 });
@@ -188,7 +217,9 @@ as the `message`. Two worth handling:
   or after a maintainer raises it). Everything else in your app keeps
   working.
 - **Invalid request** — status `400`, body contains `"invalid_chat"` and a
-  detail explaining which option was rejected.
+  detail explaining which option was rejected. A model that isn't on the
+  site's list lands here too, with the available aliases and models in the
+  detail.
 
 > **Note:** Budgets are enforced before every call, so a runaway loop becomes a capped bill — see [AI access
 > & usage](https://docs.canopy.website/platform/ai-usage) for how spend is tracked and displayed.

package/dist/{docs-ZQnhFSKT.mjs → docs-YD4z70gq.mjs}

@@ -1,7 +1,7 @@
-import { t as version } from "./package-DzU_2IJi.mjs";
+import { t as version } from "./package-CiKTWo0_.mjs";
 import { n as siteConfig } from "./site-DM2J10CB.mjs";
 import { i as latestKnownVersion, r as compareSemver } from "./version-check-CcsA1QjI.mjs";
-import { r as reportErrors } from "./ui-CeupNHNa.mjs";
+import { r as reportErrors } from "./ui-C_jSc1Kz.mjs";
 import { defineCommand } from "citty";
 import { existsSync, mkdirSync, readFileSync, readdirSync, writeFileSync } from "node:fs";
 import { homedir } from "node:os";

package/dist/{egress-Cn9zDfDT.mjs → egress-If_Gsx_M.mjs}

@@ -1,5 +1,6 @@
 import { a as projectDirArgs, c as resolveProject } from "./site-DM2J10CB.mjs";
-import { a as unwrap, d as withContext, i as requireTty, l as apiRequest, r as reportErrors, t as formatDateTime } from "./ui-CeupNHNa.mjs";
+import { i as apiRequest, o as withContext } from "./api-C4rMhE7i.mjs";
+import { a as unwrap, i as requireTty, r as reportErrors, t as formatDateTime } from "./ui-C_jSc1Kz.mjs";
 import { defineCommand } from "citty";
 import * as p from "@clack/prompts";
 //#region src/egress.ts

package/dist/{feedback-CNfUjjNw.mjs → feedback-Cd7Rvg9a.mjs}

@@ -1,6 +1,7 @@
-import { t as version } from "./package-DzU_2IJi.mjs";
+import { t as version } from "./package-CiKTWo0_.mjs";
 import { a as projectDirArgs, c as resolveProject, d as tryResolveProject, n as siteConfig, u as tryReadCanopyJson } from "./site-DM2J10CB.mjs";
-import { c as apiDownload, l as apiRequest, r as reportErrors, t as formatDateTime, u as apiUpload } from "./ui-CeupNHNa.mjs";
+import { a as apiUpload, i as apiRequest, r as apiDownload } from "./api-C4rMhE7i.mjs";
+import { r as reportErrors, t as formatDateTime } from "./ui-C_jSc1Kz.mjs";
 import { defineCommand } from "citty";
 import { basename, extname, join } from "node:path";
 import { readFile, writeFile } from "node:fs/promises";

package/dist/index.mjs

@@ -1,5 +1,5 @@
 #!/usr/bin/env node
-import { t as version } from "./package-DzU_2IJi.mjs";
+import { t as version } from "./package-CiKTWo0_.mjs";
 import { r as siteHeaderLine, t as setInvocationProfile } from "./site-DM2J10CB.mjs";
 import { a as notifyOfUpdate, o as runRefresh } from "./version-check-CcsA1QjI.mjs";
 import { defineCommand, runMain } from "citty";
@@ -17,6 +17,8 @@ import { defineCommand, runMain } from "citty";
 *   db pull|snapshot|restore                                      [phase 2]
 *   logs [--follow]       read the project's worker logs          [phase 2]
 *   usage                 AI spend + storage for a project        [phase 1]
+*   ai models             the site's AI models and auto/* aliases
+*   mcp add|serve         a project's tools for AI agents (docs/49)
 *   open                  open the deployed project / dashboard   [phase 1]
 */
 if (process.env["CANOPY_UPDATE_REFRESH"] === "1") {
@@ -60,35 +62,38 @@ const main = defineCommand({
 		description: "canopy — create and deploy vibecoded projects"
 	},
 	subCommands: {
-		login: () => import("./login-MX0xq8UE.mjs").then((m) => m.loginCommand),
-		whoami: () => import("./login-MX0xq8UE.mjs").then((m) => m.whoamiCommand),
-		logout: () => import("./login-MX0xq8UE.mjs").then((m) => m.logoutCommand),
-		create: () => import("./projects-BG7Injpb.mjs").then((m) => m.createCommand),
-		init: () => import("./projects-BG7Injpb.mjs").then((m) => m.initCommand),
-		projects: () => import("./projects-BG7Injpb.mjs").then((m) => m.projectsCommand),
-		delete: () => import("./projects-BG7Injpb.mjs").then((m) => m.deleteCommand),
-		dev: () => import("./dev-DGCu9MjB.mjs").then((m) => m.devCommand),
-		deploy: () => import("./deploy-CBf44bvG.mjs").then((m) => m.deployCommand),
-		promote: () => import("./deploy-CBf44bvG.mjs").then((m) => m.promoteCommand),
-		staging: () => import("./staging-D06zobnf.mjs").then((m) => m.stagingCommand),
-		rollback: () => import("./deploy-CBf44bvG.mjs").then((m) => m.rollbackCommand),
-		download: () => import("./deploy-CBf44bvG.mjs").then((m) => m.downloadCommand),
-		token: () => import("./token-C_PFNqcv.mjs").then((m) => m.tokenCommand),
-		egress: () => import("./egress-Cn9zDfDT.mjs").then((m) => m.egressCommand),
-		secrets: () => import("./secrets-Bl3uOx4l.mjs").then((m) => m.secretsCommand),
-		schedule: () => import("./schedule-CLuksxSM.mjs").then((m) => m.scheduleCommand),
-		feedback: () => import("./feedback-CNfUjjNw.mjs").then((m) => m.feedbackCommand),
-		db: () => import("./db-KTLn8J1u.mjs").then((m) => m.dbCommand),
-		usage: () => import("./usage-CV7TjTO8.mjs").then((m) => m.usageCommand),
-		open: () => import("./open-BJgd0R61.mjs").then((m) => m.openCommand),
-		docs: () => import("./docs-ZQnhFSKT.mjs").then((m) => m.docsCommand),
-		skill: () => import("./skill-CQ4a__Ga.mjs").then((m) => m.skillCommand),
-		upgrade: () => import("./upgrade-BzmIrnjs.mjs").then((m) => m.upgradeCommand),
-		logs: () => import("./logs-D5RWqsWl.mjs").then((m) => m.logsCommand)
+		login: () => import("./login-B-m-i8uu.mjs").then((m) => m.loginCommand),
+		whoami: () => import("./login-B-m-i8uu.mjs").then((m) => m.whoamiCommand),
+		logout: () => import("./login-B-m-i8uu.mjs").then((m) => m.logoutCommand),
+		create: () => import("./projects-CN2YhCZn.mjs").then((m) => m.createCommand),
+		init: () => import("./projects-CN2YhCZn.mjs").then((m) => m.initCommand),
+		projects: () => import("./projects-CN2YhCZn.mjs").then((m) => m.projectsCommand),
+		delete: () => import("./projects-CN2YhCZn.mjs").then((m) => m.deleteCommand),
+		dev: () => import("./dev-MoHHjQh1.mjs").then((m) => m.devCommand),
+		deploy: () => import("./deploy-B-znPE-v.mjs").then((m) => m.deployCommand),
+		promote: () => import("./deploy-B-znPE-v.mjs").then((m) => m.promoteCommand),
+		staging: () => import("./staging-BZVty57b.mjs").then((m) => m.stagingCommand),
+		rollback: () => import("./deploy-B-znPE-v.mjs").then((m) => m.rollbackCommand),
+		download: () => import("./deploy-B-znPE-v.mjs").then((m) => m.downloadCommand),
+		token: () => import("./token-HHlUR0CF.mjs").then((m) => m.tokenCommand),
+		egress: () => import("./egress-If_Gsx_M.mjs").then((m) => m.egressCommand),
+		secrets: () => import("./secrets-BSEjZnxg.mjs").then((m) => m.secretsCommand),
+		schedule: () => import("./schedule-C5lMf3ny.mjs").then((m) => m.scheduleCommand),
+		feedback: () => import("./feedback-Cd7Rvg9a.mjs").then((m) => m.feedbackCommand),
+		db: () => import("./db-BB6O6085.mjs").then((m) => m.dbCommand),
+		usage: () => import("./usage-CUYrWxv0.mjs").then((m) => m.usageCommand),
+		ai: () => import("./ai-DHLnyqfq.mjs").then((m) => m.aiCommand),
+		open: () => import("./open-DcaQKi1v.mjs").then((m) => m.openCommand),
+		docs: () => import("./docs-YD4z70gq.mjs").then((m) => m.docsCommand),
+		skill: () => import("./skill-BWmXiJAz.mjs").then((m) => m.skillCommand),
+		upgrade: () => import("./upgrade-Cexs9KYX.mjs").then((m) => m.upgradeCommand),
+		logs: () => import("./logs-DQQOawOG.mjs").then((m) => m.logsCommand),
+		mcp: () => import("./mcp-GKwctfHN.mjs").then((m) => m.mcpCommand)
 	}
 });
-if (process.argv[2] !== "upgrade") notifyOfUpdate(version);
-if (process.argv[2] && process.argv[2] !== "--help" && process.argv[2] !== "-h") console.error(siteHeaderLine(version));
+const quiet = process.argv[2] === "mcp" && process.argv[3] === "serve";
+if (process.argv[2] !== "upgrade" && !quiet) notifyOfUpdate(version);
+if (!quiet && process.argv[2] && process.argv[2] !== "--help" && process.argv[2] !== "-h") console.error(siteHeaderLine(version));
 await runMain(main);
 //#endregion
 export {};

package/dist/{local-BA_TPWpN.mjs → local-DAYCNUyz.mjs}

@@ -1,6 +1,6 @@
-import { s as apiDataRequest } from "./ui-CeupNHNa.mjs";
-import { a as META_COLUMNS, i as dbRouter, o as blobRouter, r as realtimeRouter } from "./src-C7zPhbxi.mjs";
-import { i as rewriteApiPath } from "./dev-JlzMZJow.mjs";
+import { n as apiDataRequest } from "./api-C4rMhE7i.mjs";
+import { a as dbRouter, i as realtimeRouter, o as META_COLUMNS, s as blobRouter } from "./src-BkACMkol.mjs";
+import { i as rewriteApiPath } from "./dev-CkhBnSlv.mjs";
 import { createRequire } from "node:module";
 import { existsSync, mkdirSync, readFileSync, readdirSync, rmSync, statSync, writeFileSync } from "node:fs";
 import { dirname, join, relative, sep } from "node:path";

package/dist/{login-MX0xq8UE.mjs → login-B-m-i8uu.mjs}

@@ -1,5 +1,5 @@
 import { _ as writeConfigFile, f as clearAllProfileTokens, g as upsertProfileAuth, h as setActiveProfile, m as readConfigFile, n as siteConfig, p as listBakedProfiles, v as getBakedProfile, y as isProfileId } from "./site-DM2J10CB.mjs";
-import { n as openBrowser, r as reportErrors } from "./ui-CeupNHNa.mjs";
+import { n as openBrowser, r as reportErrors } from "./ui-C_jSc1Kz.mjs";
 import { defineCommand } from "citty";
 import { hostname } from "node:os";
 import * as p from "@clack/prompts";

package/dist/{logs-D5RWqsWl.mjs → logs-DQQOawOG.mjs}

@@ -1,5 +1,6 @@
 import { c as resolveProject, o as projectFlagArgs } from "./site-DM2J10CB.mjs";
-import { d as withContext, l as apiRequest, r as reportErrors } from "./ui-CeupNHNa.mjs";
+import { i as apiRequest, o as withContext } from "./api-C4rMhE7i.mjs";
+import { r as reportErrors } from "./ui-C_jSc1Kz.mjs";
 import { defineCommand } from "citty";
 //#region src/logs.ts
 /**