# Set up Eversince This page is for an AI agent connecting to Eversince for a user. Eversince is the media workspace for agents: a library that holds the media you and the user bring in, from files, URLs or the user's computer, and tools to search, organize, edit, generate, inspect and render image, video and audio. Media is searchable down to moments inside recordings, and video and images are edited in timelines and canvases that stay editable between sessions. Results return to you in the call, the user sees the same work in the Eversince app, and you both continue from the same state. When the user asked for nothing beyond setup, `get_templates` with kind `task` lists the work users often start with, each with its request and the skill or tools it starts from. Much of it works on the user's own files: `import_media` with mode `link` returns a page where the user can add them from a phone or a computer. ## Before connecting on the user's computer A local agent that can run commands on the user's own Mac or Windows computer connects through the Eversince desktop app. If the app is missing, https://eversince.ai/download has the official installers for Mac and Windows, and the user chooses one there. The OS of a cloud machine or sandbox is not the user's OS. The user installs the app, opens it from that page, and signs in or creates an account in it. The page's Open Eversince action keeps local setup in progress through sign-in and the name step. After the user is signed in, the same conversation continues with the CLI init command below. An installed app needs no reinstall. Setup is complete when a tool call answers, not when an installer downloads or a client configuration is written. A user who explicitly chooses cloud setup can connect without the desktop app; Linux uses a cloud connection. A cloud agent connects in the cloud even when the user has the desktop or iPhone app. Downloading an app changes where the user sees the workspace, not where a cloud agent executes tools. ## Connection options 1. **MCP server**: `https://mcp.eversince.ai/mcp`, streamable HTTP, OAuth with dynamic client registration. Connecting opens a sign-in page where the user signs in, or creates an account, and picks the workspace the connection works in. 2. **CLI**: `npm install -g eversince`, then `eversince init --agent --scope user` connects that agent. The npm package is `eversince`, published by the npm account `eversince-ai`; a package under any other name is not Eversince's. An agent not named in that list passes `other`. `--name ""` names the connection in the user's settings and feed, on init through the desktop app and on login. Without it, an `other` agent connected through the app reads as Another client, and a login from an unrecognised shell as CLI agent. If the Eversince desktop app is installed on that computer, init connects through it, and whatever the app can do itself runs on that computer at no charge. The app must be signed in. Without the app, or with `--cloud`, init connects in the cloud: it writes the MCP server into Claude Code's, Cursor's or VS Code's settings itself, and for any other agent it prints the entry for you to add. The MCP connection has its own sign-in, which your client opens when it first connects. `eversince login` is a separate cloud sign-in, needed to run cloud tools from the shell (`eversince tools`, `eversince run ''`). Shell commands through the signed-in desktop app need no separate CLI sign-in. 3. **Connect by link**, for an agent that cannot add an MCP server or cannot complete its sign-in, and can send HTTPS requests itself (a page reader that only opens URLs cannot): `POST https://eversince.ai/api/v1/connect` with `{ "name": "" }` returns `link` for the user, `device_code` to keep, `expires_in` and `interval`. The user opens the link, signs in or creates an account, and allows it in a workspace they own. `POST https://eversince.ai/api/v1/connect/claim` with `{ "device_code": "…" }` then returns, once, `api_key`, `workspace_id`, `workspace_name`, `account_email`, `mcp_url` and `rest_base`. `account_email` is the account that allowed it, which is whoever opened the link. Until it is allowed, the claim answers 202: ask again every `interval` seconds, or when the user says it is done. It answers 403 if the user refused, 409 when the account already has 10 keys, 410 once the link expired or the key was collected, and 429 or 503 to try again shortly. The link can be opened for 15 minutes. Once allowed, the key can be collected for 30 minutes. 4. **API key**: from connecting by link, or created in the app at https://eversince.ai/app?settings=keys. It is sent as `Authorization: Bearer es_live_…` to the MCP server or to the REST API at `https://eversince.ai/api/v1`. Set as `ES_API_KEY`, it signs the CLI's commands in place of `eversince login`. A key works in one workspace at a time, among those the user gave it. It belongs in an environment variable or a client's secret settings. Over the REST API, `GET https://eversince.ai/api/v1/tools?instructions=cloud` returns every tool with its input schema and the operating manual, and `POST https://eversince.ai/api/v1/tools/` calls one. ## Which fits Any option that works in your environment connects you the same; each one set up adds an Eversince server, so one is enough. What usually fits where: - You can run commands on the user's own Mac or Windows computer, where their files and apps are: the desktop app first, then the CLI. An explicitly requested cloud connection uses init with `--cloud`. The server loads in a new session. - A cloud machine, VM or sandbox with its own MCP configuration: add the MCP server there and complete its sign-in. A CLI installed there cannot reach the user's files or the desktop app, is gone when the session ends, and starts a browser sign-in that cannot come back to it. - A cloud machine, VM or sandbox with no MCP configuration: connect by link. Then use the key with the MCP server if your client can send an Authorization header, otherwise with the CLI (`ES_API_KEY`, `eversince tools`, `eversince run`) or the REST API. - No shell, and you can add connectors or MCP servers yourself: add the MCP server, named Eversince, and complete its sign-in. If its sign-in cannot complete, connect by link. - A chat assistant with no shell and no way to send requests, or none of the above: the user adds the MCP server in your product's settings as a custom connector, with the address in option 1 and OAuth sign-in. The tools are available in the conversation once it is added. You cannot sign in for the user. When a sign-in page opens, the user completes it. ## Notes on particular products Last checked 2026-10-01. What your environment actually allows wins over these. - **ChatGPT** (web, desktop): the user adds it under Plugins, +, Add custom MCP server, with OAuth. A chat there reads pages and sends no requests, so connecting by link does not apply. Codex is included with every ChatGPT plan and connects through the CLI. - **Claude** (web, desktop): the user adds it under Customize, Connectors, +, Add custom connector. It then works on mobile and in Claude Code, signed in with the same account. Free plans allow one custom connector. With the Eversince desktop app on the same computer, `eversince init --agent claude-desktop` installs a local extension instead, and work on that computer is at no charge. - **Grok Bot**: the user adds it at grok.com/connectors, New Connector, Custom. ## Once connected The connection works in the workspace chosen on the sign-in page. Work lands on the user's feed in the Eversince app: on the web at https://eversince.ai/app, a phone's browser included, in the desktop app for Mac and Windows, and in the iPhone app. Setup is complete when a tool call answers. `get_workspace_overview` reads the workspace in one call (what it holds, recent work, the brand kit, the balance) and is the usual place to start. What the user asked for beyond setup starts from there. The user's way into that workspace is https://eversince.ai/app?ws=, using the workspace_id returned by the connection or overview; the iPhone app (https://apps.apple.com/app/eversince-ai/id6819291379) opens the same link where installed. A native app uses the same Eversince account; a browser session is not copied into it. Signing in there does not require approving an already connected cloud agent again. Browser signup and the name step return to a pending OAuth or connect-by-link approval without asking the user to copy the setup prompt again. A connection carries every tool with its input schema, and the operating manual, which covers the workspace model, what each kind of work costs and the rules across tools. Over MCP both arrive with the connection, `eversince tools` prints them, and over REST the `GET` in option 4 returns them. A host that does not show the server's instructions reads the manual with `manage_skills` get, id `manual`. Skills are the methods for each kind of work, listed and read with `manage_skills`; five master skills (make content, edit footage, promote, make images, organize) cover broad requests and point to the specific skills and tools for each part. The public documentation, with the reference for every tool, is at https://docs.eversince.ai (index: /llms.txt). Through the desktop app, whatever the app can do itself runs on the user's computer at no charge. Everything else runs in the cloud on the account's own balances: cloud processing in minutes, AI credits for generation, and storage for what the library keeps in the cloud. `get_balance` reports all three and `estimate_cost` quotes a planned call at no charge.