What a connected app can do
- List the brands you can work in and pick one. Each call can also name its brand, so several chats in the same app can work in different brands at once.
- Create a brand, personal or business, and set it up from a website, public profiles, your own description, or files in your project.
- Read brand context: voice, positioning, audience, colors, typography, logos, brand images, visual styles, and brand documents, or just the parts that apply to one piece of work.
- Search and read the brand’s documents and brand images.
- Start marketing work in the brand, in your words, such as research, campaigns, posts, articles, and SEO.
- Follow a thread’s progress and latest update, and reply in it.
- Set up routines: recurring work Daoco runs on a schedule in the brand, such as “every two hours, draft one new post idea”, that keeps running when the app is closed.
- List recent threads, drafts, and what is waiting on a person, each with a link into Daoco.
- Post to the brand’s connected social accounts directly: save a post exactly as written, then publish it now or schedule it, without a Daoco chat and without using workspace tokens.
- Use the brand’s other connected apps directly, such as a CRM, store, or workspace tool: list their tools and call one, without a Daoco chat.
- Read the brand’s website improvements and record what it did with them, and read the brand’s content ideas: what Daoco suggests making next and how past ideas did.
Posting and approvals
create_post saves the post your app wrote as a draft, unchanged, to the accounts you name. schedule_post publishes it now or at a time you give. Your brand’s approval setting (Settings > Approvals) decides whether a person has to approve first, the same way it does for a post Daoco writes in a chat:
- Full autonomy: the post is scheduled, or starts publishing, straight away.
- Work autonomously: publishing now waits for approval. A schedule more than an hour away is set straight away, because it can still be cancelled, unless publishing is set to ask under Approvals or the account has its own approval setting.
- Ask before acting: every publish and schedule waits for approval.
update_post cannot change it. An approval covers the post as it was when the person approved it: if the post changes before Daoco publishes or schedules it, or the brand’s settings stop allowing it, nothing goes out and the approval has to be asked for again. cancel_post takes a scheduled post off the schedule, or withdraws a request your app made that is still waiting for approval, and leaves the draft. A post that is published or publishing cannot be cancelled from the app.
Some platforms need settings before a post can go out, given in platformOptions: a privacy level for TikTok (who can see the post), a title for a YouTube Short, a title and board for a Pinterest Pin, and a title and subreddit for Reddit. TikTok also takes whether to allow comments, Duet, and Stitch, branded-content disclosure, and sending the video to the TikTok app’s inbox instead of posting it; a Short also takes its visibility. These settings are part of what an approval covers. When you attach a video, Daoco measures its length and frame size when the post is saved, so the platforms’ length and orientation rules can be checked; you don’t send those numbers. create_post, update_post, and get_post list anything that still stops the post from going out as blockingIssues. They also return brandCheck, Daoco’s Brand check of the saved post. It is checking for a few seconds after a save; get_post with waitSeconds answers when the check finishes.
Use start_work when you want Daoco to write, plan, or make the media for posts; use create_post when your app already has the exact post.
Connected apps
list_apps lists the brand’s connected apps, and with an appId lists that app’s tools, their inputs, and whether each call reads, writes, publishes, or deletes. call_app_tool runs one tool with the arguments your app gives. A read runs straight away with no approval, and its answer goes to your app, including private data the connected account holds, such as contacts, orders, or messages. A write, publish, or delete follows the same approval setting as posts, and any setting you gave that connection in Daoco: when it allows the change, it runs straight away; otherwise it waits for you to approve it in Daoco or from your personal assistant, and approving runs exactly the arguments you were shown. An approved call does not run if, before it does, the connection was turned off or set to stricter access, or reconnected to a different account. get_app_call reports the result, and cancel_app_call withdraws a call still waiting for approval. Changes to Google Ads campaigns are not available here; ask Daoco with start_work.
Website improvements
The fixes Daoco’s site audits and AI visibility checks record are on the Improve tab of your Website, and your app sees the same list.list_improvements lists them with what was found, in plain words, their status, whether anyone is on them (progress), and the metric each one is watched by. get_improvement gives one improvement’s detail: what was found and the measured signal behind it, the suggested change, and when it counts as done. A coding agent such as Claude Code can make the change in your own codebase, then call update_improvement with done. It can record several at once with ids (up to 20), and add a note and link when it starts (in_progress) or finishes (done), such as the branch or pull request it is working in.
list_improvements returns 20 improvements by default and up to 60. total says how many match, and nextCursor gives the next page: pass it back as cursor with the same filters. Each improvement says when a check last recorded it (observedAt) and first did (firstObservedAt). When the page a fix is about now redirects or no longer exists, Daoco closes the fix and leaves it out, unless someone took it on; the Improve tab lists it under Closed automatically with the reason. A fix someone took on (an app, a Daoco conversation, or marked done) stays, with pageNow saying where the page now leads (redirectsTo) or what it answers (statusCode), and update_improvement still works on it.
Your app works on improvements only when you ask it; Daoco cannot start it. Choose Queue for AI app on an improvement in Daoco to set it aside, then ask your app something like “Work on my queued Daoco website improvements.” list_improvements with status queued returns exactly those, and get_brand_context says how many are waiting. Only Daoco’s record changes: for something a page audit measures, such as a title, structured data, or an FAQ section, Daoco confirms the fix when a later audit of that page measures it fixed, exactly as for a change made any other way, and the improvement waits until such an audit runs (the weekly review re-audits pages). Other fixes count as live once marked done, unless the check that found them records them again; an AI visibility finding read from the AI answers themselves is judged by the AI mention rate instead. Daoco then measures the result. Improvements you dismissed in Daoco are not shown to the app, and neither are judgments that ask for no change (a page that already answers its query, for example). To have Daoco make the change through your CMS or GitHub, use start_work with /fix-site and the improvement’s id. A queued improvement a Daoco thread takes on (with /fix-site, the weekly review, or your assistant) leaves the app’s queue and shows as in progress in that thread.
Content ideas
list_content_ideas reads the same ideas the What to make next cards show in Daoco: pages and posts Daoco suggested from your social results, Search Console, AI visibility, keyword demand, and competitors’ rankings (see What to make next). Each idea has the evidence behind it, a score, its status, and startWork: the request to pass to start_work to make it, such as /write-page best running shoes for flat feet. view: "history" lists ideas started or made, with how each made post did against the account’s usual engagement rate and each published page’s watched metric; whatWorked sums that up by format and platform. total counts the ideas the view reads, not every idea the brand has: open reads the 20 strongest suggested pages and the 50 strongest open post ideas, history the 200 most recently updated post ideas and up to 50 pages from the last year, and dismissed up to 200 post ideas and 20 pages. Reading ideas records nothing; Daoco records ideas when it plans them, with /make-next or /ideas.
Connecting accounts from your app
When a post needs an account the brand hasn’t connected, or one that needs reconnecting, your app can askget_connect_link for a link and give it to you. Opening it asks you to sign in to Daoco and shows the account, brand, and workspace it connects. Confirm, and it goes to the platform’s sign-in, such as LinkedIn’s, and the account is connected to that brand. Bluesky and Reddit, which connect with an app password, open their setup on the Connections page. Integrations that need an API key, a site address, or an admin open their setup on the Connections page instead, so a key never travels in a link. Anyone with access to the brand can use the link, after signing in, for 24 hours; someone outside the workspace gets nothing from it. Your app can call list_channels (or list_apps for an integration) with waitSeconds to see the account as soon as it is connected. An integration can be connected only by the person who created the brand or a workspace admin; for anyone else the link opens the Connections page.
Let an app publish without asking
A workspace owner or admin can let one app or API key publish without a person approving first. On the Assistants page, open the app and turn on Let this app publish without asking. It is off by default, it applies only to that app or key, and the page shows who turned it on and when. With it on,schedule_post and call_app_tool skip approval when the call passes approval: "skip". A call without it follows your brand’s approval setting as before. With the setting off, approval: "skip" is ignored: the call follows your approval setting, and the result says approval was needed because the app is not allowed to publish without asking.
Some actions still need a person every time, with or without the setting:
- Deleting or removing anything in a connected app.
- Email: any change in an email or newsletter app (Gmail, Outlook, Ghost, beehiiv, Mailchimp, Klaviyo, SMTP) other than saving a draft, and anything in an outreach tool such as Smartlead.
- Sends and changes over Daoco’s limits, such as more than 500 recipients or more than 100 records at once.
- Changes through a connection set to always ask, or to approve each change one at a time.
start_work or continue_work: those threads follow your approval setting. Turning the app off or deleting it removes the setting, and allowing it again does not bring it back. It also stops applying if the app moves to a brand in another workspace, the brand moves to another workspace, the app signs in again as a new registration, or the person who turned it on is no longer an owner or admin. A call already under way stops too, as long as it has not yet published or changed anything.
Connect an app
The server address ishttps://backend.daoco.org/mcp. Selecting a brand requires a free assistant slot; starting or continuing work uses that workspace’s tokens.
Start from Daoco or from the app. In Daoco, open the Assistants page, choose Create on Connected assistant, pick your app, and follow its steps; the page shows when the app has signed in. In the app, add the address as a custom MCP connection and sign in.
Custom connections and public directory listings have separate availability. These instructions describe adding Daoco by address. Use a directory’s Daoco listing only when it is available in your app; supporting MCP does not guarantee that an app lists Daoco or supports our OAuth flow. Your app’s plan and administrator settings may restrict custom connections.
- Claude
- ChatGPT
- Grok
- Muse
- Claude Code
- Codex
In Claude, open Settings > Connectors, choose Add custom connector, and paste the address. Sign in with your Daoco account when asked.
mcp.json. These are client setup references; Daoco compatibility has not yet been verified end to end for these clients.
This address connects directly to Daoco’s backend. If you previously added a convex.site address or https://daoco.org/mcp, update the server address in your app and sign in again. OAuth tokens are issued for the configured server address.
Connect a script with an API key
Scripts and agent runs that cannot complete a sign-in can use an API key. On the Assistants page, choose New connected assistant > API key, give the key a name, and pick the brand it starts in. Daoco shows the key once; copy it then. Send it on every request tohttps://backend.daoco.org/mcp:
use_brand. It acts as you, with your workspace access, and has the same tools and limits as an app that signs in, so approvals still happen in Daoco. Treat it like a password: keep it in a secret store, not in code you share. Turn off stops a key until you allow it again; Delete app removes it for good, and anything using it stops working at once. A deleted or unknown key gets a 401 with no sign-in challenge. If Daoco switches API keys off, requests get a 503 and the key is not checked; keep it and retry later.
Start from an AI app
You can start using Daoco from an AI app without opening Daoco first.- Add Daoco to your app and sign in. If you have no Daoco account, signing in creates one with your verified email.
- Ask the app to set up your brand. Say whether it is a personal brand (you are the brand: a creator, founder, or expert) or a business, and give it your website, public profile links such as LinkedIn or Instagram, a description in your own words, or your brand files (see below). For a personal brand you can add what you are after, such as clients, speaking, an audience, or brand deals.
- Daoco creates the brand and selects it for the app. If you administer no workspace, it goes in a new workspace of your own on the free Starter trial, with no card needed. If you administer more than one, the app asks which.
- Daoco researches the brand, which takes a few minutes.
- The app shows you what Daoco set up, decisions first: whether this is your brand (name, what you do, the logo as an image, colors, and fonts), the likely competitors it found, the questions research could not answer, and the Daoco Looks it found in your images (visual styles for the images Daoco makes, each with example links, for you to keep or not). Who it is for and how you sound follow. Ask the app to remove or add a competitor (a name, and its website if you know it; Daoco looks it up otherwise), or to confirm the list. Prices, fees, numbers, and certifications you mention in your answers are saved as facts and written into your documents as you gave them. When one disagrees with your website or a file you shared (your site says $19 and you said $15), the app asks which is right, at most three times, and a later competitor search never changes a list you have answered about. Nothing waits on your approval: Daoco starts writing about a minute after the profile is drafted, or as soon as you confirm. Until then, fixes, answers, kept looks, and competitor changes go into the brand’s documents; fixes, answers, and competitor changes are still taken after that, for a day after you create the brand, and each rewrites the documents it affects shortly after you stop answering. A document you edited yourself is never overwritten. To skip a question, give it an empty answer. If you have not decided by then, the profile stays as drafted, the competitors found stay (minus any you removed), and looks you did not keep stay suggestions. For a day after you create the brand, tell the app anything that is off, such as an old logo or the wrong colors, and Daoco changes the brand, even after setup has finished. To swap the logo, give the app the image or a link to it.
- Daoco writes the brand’s summary, voice, and documents from the profile and your answers, and researches each competitor with a website, which joins the brand’s watch list. Changes made after setup has finished go through brand updates: removing a competitor then takes it off the watch list, and the documents are not rewritten.
- Once setup finishes, the app can offer the first work Daoco suggests for the brand, such as a point-of-view post or a bio kit. The suggestions stay until an AI app starts work in the brand or you start your first deliverable from onboarding in Daoco.
Start from your own files
In an app that can read your files, such as Claude Code, Codex, or Cursor, you can set up the brand from material you already keep in a project folder or repository: brand guidelines, voice and positioning notes, product or audience documents, an about page, or a Daoco brand export. Ask the app to set up your brand in Daoco from those files. The app reads them and sends their contents withcreate_brand; Daoco cannot open files on your computer by itself.
- Daoco reads Markdown, text, CSV, TSV, JSON, and YAML files, up to 40 files of 200,000 characters each, within a 1 MB request. PDFs and images are not read at setup; add them in Daoco.
- Each file becomes one of the brand’s shared files, read the same way as files you upload during setup in Daoco. You can give files with or without a website: with only files, Daoco drafts the profile from them.
- Files you send are read even when they look thin; Daoco sets one aside only when it is plainly not about the brand, looks like a legal document (a contract, NDA, service agreement, privacy policy, or terms of service, so its terms never become brand claims; the brand’s own social media policy or usage guidelines are read), or has no readable text.
review_brand_setupsays how many files were read and names each one that was not, with the reason; file names arrive marked as file content. - A Daoco brand export (the folder Export brand downloads, marked by its
daoco.json) is read as the brand itself: its brand documents become the new brand’s documents, so setup writes only the missing ones; its visual styles arrive as suggested styles and its competitors with watching off. Send one brand’s folder at a time; a workspace export holds several. With only the export, and no website or profiles, setup starts from the export’s own profile, logo, and Brand Summary with no research, so the review is ready in seconds and Daoco starts writing about a minute later unless you confirm sooner (the Brand Summary is kept only when the new brand has the export’s name, and a correction you make before then writes a new one from your corrected profile); give a website or profiles as well and Daoco researches the brand as usual, reading the export as evidence. Other files you send with the export, ones an export does not write, are read like any other files. When setup starts from the export alone, they are saved as the brand’s shared files for later work instead, and the result lists them insharedFiles. With an export, a file Daoco does not read at setup (an image, code, or an empty file) is left out and listed inskippedFilesinstead of refusing the call. The export’s library files (itsassets/folder) are listed there too: Daoco adds them only when you import the export zip in Daoco. To start a brand from an export in Daoco itself, drop the zip under the link on the first step of setup. - A
SKILL.md, with the files in itsreferences/folder, is imported as a draft skill. An owner or admin of the workspace reviews and publishes it on the Skills page before Daoco uses it. A skill whose command is already taken in the workspace is not imported, and the app is told why.
Follow marketing work
Ask your app to list your brands, select the one you want, and hand Daoco your request. For example:- “Use my Acme brand and tell me its voice and target audience.”
- “Draft three LinkedIn posts for Acme’s product launch and keep me updated.”
- “Make the second post shorter and show me what needs review.”
Routines
A routine is recurring work your app sets up in Daoco withcreate_routine: a name, the instructions each run follows, and a schedule, which is one of every so many minutes (intervalMinutes, at least 30), a cron expression (cron, read in timezone, at most every 30 minutes), or a single time (runAt). Daoco runs it on that schedule whether or not your app is open. Each run starts a new thread in the brand, exactly as start_work does: it is your chat, owned by you, it uses your workspace’s tokens like any other work, and its drafts and approvals follow your approval setting, decided in Daoco or from your personal assistant. Each run is also told what the routine’s last few runs reported and which of the brand’s posts are scheduled, waiting for approval, or recently published, so it can pick up where the last run stopped and not repeat a post. A run that is still working when the next one is due skips that one.
Your app follows the runs like any other work: list_routines lists the app’s routines with their next run and their latest runs, each with its threadId, and list_work, get_work_status, and read_work read those threads. update_routine changes a routine’s name, instructions, schedule, or timezone, or pauses it and turns it back on. delete_routine stops it for good; the threads it made stay. A one-time routine runs once and is then no longer listed, so update_routine and delete_routine no longer find it; its run’s thread is in list_work. An app can hold up to 25 routines, and they use no extra assistant slot.
Routines belong to the app that made them and run as the person who connected it. Turning the app off, removing it from the workspace, or leaving the workspace pauses its routines there, and deleting the app, or moving the brand to another workspace, stops them for good (the threads they made stay); allowing the app again leaves them paused until they are turned back on. A routine also pauses when its brand is no longer available to you, or after its last three runs failed. Its app’s page on the Assistants page lists its routines, where the person who connected it or a workspace admin can pause, turn on, or delete each one.
If your app still lists list_recent_work, find_drafts, list_needs_you, show_visual, or get_asset, refresh its tool list or reconnect. Those tools have been replaced by list_work and get_output.
MCP tool reference
The server exposes these tools to authenticated clients. All tools exceptlist_brands, use_brand, and create_brand work in a brand and take an optional brandId. A call works in the brand it names; without one, in the brand of the threadId, postId, callId, or routineId it names; otherwise in the app’s selected brand. Naming a brand never changes the selection, so separate chats in one app can each work in their own brand. Working in a brand of another workspace takes one of that workspace’s assistant slots unless the app already holds one there, and list_brands marks the brands where it would. A workspace slot taken this way is freed after two weeks without work there, or from that workspace’s Assistants page with Remove from this workspace. When you leave a workspace, your apps free their slots there, and an app whose selected brand was in it has no brand selected until it picks another. An API key bound to a brand there is turned off and loses that brand. Reads return brand data or work summaries; writes create or select a brand, change a new brand’s setup review, or create a message in a durable Daoco thread. Every successful result from a tool that works in a brand names that brand as brand: an object with its id and name, except where a tool already named the brand before (list_channels, the post tools, list_apps and the app call tools, get_connect_link, list_work, start_work, and continue_work): there brand is the brand’s name as a string, as it always has been.
Each tool also tells your app how it behaves. Read tools are marked read-only.
review_brand_setup, stop_work, schedule_post, call_app_tool, and delete_routine are marked destructive, because a confirmed setup, a stopped run, a published post, a change in a connected app, and a deleted routine cannot be undone, so some apps ask you before calling them. create_brand, start_work, continue_work, schedule_post, list_apps, call_app_tool, and create_routine are marked as reaching outside Daoco, because the work they start can research the web and read your connected apps.
Poll status no more than once every 20 seconds, or call get_work_status with waitSeconds instead of polling.
The same goes for anything waiting on a person. A post that is awaiting_approval or publishing, a connected-app call that is awaiting_approval, and an account the person is connecting in Daoco all change only when the person acts in Daoco or the platform answers. Call get_post, get_app_call, or list_channels with waitSeconds: 30, and call again while waitTimedOut is true and the status has not changed, rather than asking the person whether it is done. Without waitSeconds, read no more than once every 20 seconds. Stop after 10 waits in a row that end with waitTimedOut (about five minutes), or sooner when the person says to, and read again only when the person asks: an approval can wait for hours, and the post or call still goes ahead when it is answered. Each app has at most 3 waits open at once, across every tool; a wait past that answers at once with waitSkipped: true instead of waiting, so treat it like a read without waitSeconds and wait 20 seconds before the next one. When the work finishes, read_work returns the full result. stop_work stops a run; calling it again on a stopped thread changes nothing. A thread someone stopped, with stop_work or Stop in Daoco, has the status stopped in get_work_status and list_work, not failed.
start_work, continue_work, create_brand, create_post, call_app_tool, and create_routine accept an optional clientRequestId, up to 200 characters. Use a new value for each request and the same value when retrying it. A repeat of a key from the same app within 24 hours returns the first call’s result, marked replayed: true, instead of starting work or sending the message again. A key reused for a different request is refused. If the first call is still running, the repeat returns an error; retry after a few seconds. For start_work, continue_work, create_post, and call_app_tool, a refusal such as insufficient tokens is replayed too, so a new attempt needs a new key. create_brand and create_routine record a key only when they create something: a refused call, such as one that needs a workspaceId or one past the routine limit, records nothing and can be retried with the same key. A repeat of create_brand returns that brand. A repeat of create_routine returns that routine, or a not-found error once the routine has been deleted or, for a one-time routine, has run. create_post keeps a key for as long as the post exists: a repeat returns the same post. Without a key, each call starts new work, sends a new message, saves a new post, or creates a new routine, so after a dropped connection check recent work and the thread’s status before retrying.
Sending images and videos with work
start_work and continue_work take files: up to 20 images (PNG, JPEG, WebP, or GIF, up to 20 MB each) or videos (MP4, MOV, or WebM, up to 2 GB each). PDFs and other documents can’t be attached from an AI app; attach them in a Daoco chat. Daoco attaches them to the message the same way it attaches a file uploaded in its own chat, so the specialists working on the message see them, for example as a style reference for new images, or a finished cut from CapCut to post. Each entry is one of:
- An
uploadIdfromupload_file. Send the file’s bytes as the body of an HTTP PUT to the returneduploadUrl, for examplecurl -T reference.png "<uploadUrl>", then pass theuploadId. For a video, name the file with its extension, pass its size in bytes assizeBytes, and send the returned headers, for examplecurl -T launch.mp4 -H "Content-Type: video/mp4" -H "If-None-Match: *" "<uploadUrl>". Each URL expires after an hour. This suits apps that can read local files, such as Claude Code and Codex. - A public
https://link to an image. Daoco downloads it and refuses links to private networks. Videos go throughupload_file.
files with a request that starts with /content-dump, for example /content-dump photos from the weekend market. Daoco picks the best shots, suggests platforms, fits each to its shape, writes the captions, and saves drafts for review in Daoco. Nothing is published or scheduled.
Daoco cannot read a file path on your computer. If a request mentions one and has no files, the result includes a notice saying the file was not sent. A clientRequestId covers the files too: retrying the same key with different files is refused.
get_brand_context and search_brand read what Daoco already stores and use no workspace tokens. With task, Daoco picks the visual style the same way it does for its own images: a style the task names, then the style set for a platform it names, then the brand default. It does not start work or write anything. Brand images in search_brand are the images marked as core brand assets; the rest of the library is not searched.
Charts, reports, images, and videos Daoco makes in a thread are returned by get_output, and get_work_status lists them with their IDs. Apps that support the MCP Apps extension can show a chart or report in a sandboxed frame, and the same frame shows an image’s preview. Rendering in individual clients is not yet verified. Asking for changes there sends your request to the chat, and your app passes it to Daoco in that thread. Other apps get a PNG image link for a chart. The visual’s page cannot call Daoco’s tools. Links in Daoco’s written answers are replaced with [stored asset], so use get_output to get the files. An image result carries an MCP image block for apps that display or read images, and every image or video result carries a resource link to the original file. Whether an app shows the image or only the link depends on the app.
Tool results include text and, on success, structured data. A tool result with isError: true reports a failed call. Missing or expired authentication returns HTTP 401 and requires sign-in; a disallowed browser origin returns 403. A missing account, lost workspace access, a turned-off app, unavailable assistant slots, or insufficient tokens can block a tool call. Follow the returned explanation or Daoco link before retrying.
Authentication and data access
OAuth sign-in uses WorkOS AuthKit. The protected resource metadata is athttps://backend.daoco.org/.well-known/oauth-protected-resource/mcp. The advertised scopes are openid, profile, email, and offline_access. Workspace membership determines which brands the person can access; these scopes do not grant access to other workspaces.
Connected apps receive the brand context, draft content, and thread information requested through their tools. Work requests and follow-ups are stored as ordinary Daoco chat messages. Review your AI app’s data policies before sharing brand documents or customer information. Turning off an app stops its MCP access; deleting its connection does not delete marketing work already created in Daoco.
See and turn off connected apps
Connected assistants appear on the Assistants page in their own Connected assistants section, since each app picks the brand it works in. An app appears after its first request, however you added it. Open an app for its page. Until the app works in a brand, the page keeps the setup steps and says what it is waiting for. After that it shows the brand the app works in, what needs you, the routines it set up (with a switch to pause or turn on each, and Delete routine), and the threads it started or continued, each with its status: Working, Waiting on you, Needs you, Done, Failed, Timed out when a run ran out of time, or Stopped when a person stopped it from the app or in Daoco; Connection holds who connected it, when it last worked, and the steps to add it again. Ask the app to list your brands and pick one. The person who connected it, or a workspace admin, can switch its brand on its page. Turn off stops the app at once and frees its slot; Allow again restores it if a slot is free. Delete app (under the ⋯ menu) removes it and frees its slot, but if Daoco is still added in the app, it comes back the next time you use it there. The person who connected an app, or a workspace admin, can turn it off or delete it. An app whose selected brand is in another workspace, but which also worked in a brand of yours, shows here as Also working here and uses one of your slots. Remove from this workspace on its page frees that slot without turning the app off elsewhere; the person who connected it or a workspace admin can do this, and the slot also frees itself after two weeks without work here. An app loses access to a brand as soon as you leave its workspace.Reading the public website
Agents can request the homepage athttps://daoco.org/ with
Accept: text/markdown to read its workflow, pricing, and FAQ as Markdown.
Requests for HTML still receive the normal website. Markdown responses include
Vary: Accept and are not cached. Unknown top-level page URLs requested as Markdown return
HTTP 404 with links to the documentation, site index, and sitemap. Missing pages
within existing sections keep that section’s normal error handling.
The public site index and
full product profile are available without signing in.
HTTP discovery and errors
The OpenAPI description lists the public website endpoints and the canonical MCP JSON-RPC transport. API discovery links to both. The unauthenticated/api/health endpoint reports whether the web
server is responding, not whether every connected service is healthy.
MCP still supplies its tool schemas through tools/list and uses its existing
OAuth and JSON-RPC error contracts. The OpenAPI description does not create REST
endpoints for marketing work. Connect to https://backend.daoco.org/mcp, not the
older dashboard alias.
Dashboard API requests without a session return HTTP 401 with
application/problem+json, a stable code, an explanation in detail, and a
resolution hint. Unmatched API routes return the same format with HTTP 404 once
the request has passed any applicable session check. Browser page navigation
still redirects to sign-in. Existing API handlers retain their own JSON errors.