Connectors and skills

Connectors give your bots tools. Each one is an MCP server: GitHub, Linear, Notion, a database, or anything else that speaks the Model Context Protocol. Skills are written instructions a bot follows for a kind of task, such as a release checklist or a house style for reports.

You add both in Extensions. They install on the computer, so every bot on that computer can use them, and each bot can turn any of them off.

Open Extensions

  • Desktop app: click Extensions at the bottom of the sidebar. With several computers, click the Extensions button on that computer's heading in the sidebar; each computer has its own.
  • iPhone and Android: tap the connection status at the top of the Bots tab (such as 1 connected), choose Manage computers, and tap Extensions on the computer. With several computers, the Extensions button on a computer's heading in the Bots tab opens it too.
  • Terminal UI: press A, or pick "Extensions: agents, connectors, skills…" from the action list. Tab switches between Agents, Connectors and Skills.

The header names the computer everything installs on. Credentials sits on that same row, beside the name. Under the open section, one line says where those items come from and that they install on this computer. Once you've added something, a count under it ("3 installed", Connectors & skills) opens your library, where you can sign in to a connector or remove one.

Add a connector from the registry

Extensions opens on one section at a time: Agents, Connectors, or Skills. On the desktop they are a row of chips under the search field; on a wide phone they are a side list, and on a narrow phone tabs along the top. Each section is one list you scroll. The Connectors list starts with the first connectors from the official MCP Registry, and more load as you scroll down to the end; if that fails, Try again at the end of the list asks again. Search by service name. In the terminal, Tab switches sections, and moving down the Connectors list loads more as you near its end.

  1. Pick a connector. The Add connector sheet opens.
  2. If it can run more than one way, choose under Runs: hosted by the service, or on your computer as a package.
  3. Fill in any Setup fields it asks for, such as an API key. Keys are saved on that computer only.
  4. Press Add. Codync installs it, opens the service's sign-in page if it has one, and checks that the connector starts and lists its tools.

Tapping a connector only opens its sheet; nothing installs until you press Add. Pressing Add on the row itself installs a connector with nothing to fill in right away.

Packages run with their own runtime, which must be installed on the computer: npm packages need Node.js (npx), PyPI packages need uv (uvx), container images need Docker, and NuGet packages need dnx. Codync installs the exact version the registry lists.

MCPB bundles and packages that serve HTTP locally aren't listed. Add those as a custom connector.

If the check fails because a key was wrong, type the correct key in the same sheet and press Add again: Codync replaces the connector it just saved with the new values. Pressing Add with nothing typed checks the saved connector again, for example after you finished signing in. You can also change a saved key later under Extensions → Credentials → Saved connections.

Sign in to a hosted connector

Some hosted connectors ask you to sign in with your account on that service. The sign-in page opens right after you add it:

  • On the iPhone, it opens in the system sign-in sheet; on Android, in your browser. A few services accept a sign-in only from the computer; for those, the phone says so, and you sign in from the desktop app or the terminal on the computer instead.
  • In the desktop app, it opens in your browser on the computer.

The tokens stay on the computer and are refreshed when they expire. If the service can't be reached for a refresh, the connector stays signed in and Codync tries again on its next use; you only need to sign in again when the service ends the sign-in. Until someone signs in, bots don't get the connector, and your library shows a Sign in button for it.

GitHub signs in with a short code instead. Codync shows the code; press Copy and open, paste the code on the GitHub page that opens, and authorize Codync. The connector is ready as soon as GitHub confirms. In the terminal, the code is copied and shown at the bottom while GitHub's page opens, or next to the page's address when the terminal can't open a browser.

If you added GitHub earlier with a personal access token, your library shows Sign in instead of using a saved token next to it. Sign in there and Codync switches the connector over and deletes the saved token, so it no longer stops working when the token expires. If one of your GitHub organizations restricts OAuth app access, an owner of that organization approves Codync first (on GitHub, in the organization's third-party access settings); until then bots can't see that organization's repositories, so keep the token if you need them.

A few services don't offer any sign-in Codync can use. For those, add the server as a custom connector with its URL and a token header.

Add a custom connector

In the Connectors section, choose Custom connector, above the registry list, then one of:

  • Command: a command line that starts an MCP server. It runs on the computer in the bot's project folder, and quotes work like in a shell. Put keys under Environment, one KEY=value per line.
  • URL: a hosted MCP server. Add headers under Headers, one Name: value per line, or leave them empty if the service has you sign in.
  • Config: paste an MCP config from a README or another app (Claude, Cursor, VS Code), such as an mcpServers block. Every server in it is added. If one isn't valid, for example its args aren't all text, none are, and the error names that server.

Everything you enter is saved on that computer only. A command or URL can't reuse the name of a connector you already have: Codync names that connector and changes nothing, so you can sign in to it or remove it in your library, or choose another name. Pasting a config still adds every entry.

Apps through Composio

Composio is in the featured connectors. Add it and sign in to your Composio account; no API key is needed. Your bots can then find and connect apps through Composio's own tools.

If you'd rather use a Composio consumer key, add https://connect.composio.dev/mcp as a custom connector with the header x-consumer-api-key: ck_….

When a bot asks you

A bot can search the registry and ask you to connect something. The request shows up as a card in the chat:

  • Connect a connector: press Connect. The setup and sign-in open as if you'd picked it in Extensions. When it's done, the connector is turned on for that bot and the bot is told.
  • Provide a secret: the card names the connector and the field the bot needs. Type the value (or a 1Password op:// reference) and press Connect. The value goes to that connector, is saved on the computer, and never appears in the conversation or reaches the bot.

The cards themselves hold no credentials. You can answer them in the iPhone app, the desktop app, or the terminal UI (pick the card and press ↵). A request can be answered for 24 hours; after that, ask the bot for a new one.

Where credentials are kept

  • Keys, headers and sign-in tokens are encrypted in Codync's database on the computer. The encryption key is held by the system: Keychain on macOS, Credential Manager on Windows, and on Linux a Secret Service keyring (such as GNOME Keyring) that is unlocked in your user session.
  • Codync never falls back to plain text. If the keyring is locked or missing, you get an error instead.
  • Once unlocked, the key stays in the host's memory until the host stops. Locking the keyring afterwards doesn't remove it; stop the host to do that.
  • Your phone sends values to the computer over its end-to-end encrypted connection. Lists in the apps show only whether a value is set.
  • Bots reach connectors through Codync on the computer, which adds the saved values when the connector runs, so the values aren't handed to the agent. This isn't a sandbox: programs running as your user on that computer, including the agent's own shell commands, aren't kept away from them.
  • Credentials don't sync between computers. A copy of Codync's data moved to another machine can't decrypt them, so back up the system keychain along with it.
  • If they can't be read anymore (bots say the key to this computer's saved credentials is missing, as after moving to a new computer without the old keychain), open Codync's Settings → Computers on that computer and, under Reset at the bottom, choose Reset next to Saved credentials and confirm twice, or run Reset saved credentials… from the terminal app's command palette. It deletes the keys, connectors, connected apps, voice keys and saved logins Codync kept there; enter them again afterwards. Agents signed in on their own stay signed in.

Extensions → Credentials shows where values are stored, lets you replace a connection's saved values (they stay hidden; bots that use the connection switch to the new values on their next turn and carry on with the conversation), and connects 1Password: paste a service account token and use op://vault/item/field references in any credential field. Give that service account access to one dedicated vault only. The 1Password CLI must be installed on the computer.

Turn connectors and skills on or off per bot

A connector or skill you add is turned on for every bot, and new bots start with all of them on (without a credential store, as on a headless Linux server, new bots start with no connectors). To turn one off for a bot, open the bot and choose Settings. Each switch there is one Codync passes on.

  • Desktop app: click the bot's name at the top of the chat, then Settings.
  • iPhone and Android: tap the bot's name at the top of the chat, then Settings.
  • Terminal UI: press e on the bot. The settings form has the same switches.

The bot picks up the change on its next turn, in the same conversation.

On a Codex bot, a connector named like an MCP server in your own Codex config (for example notion, after codex mcp add notion) reaches the bot as codync-notion, so the bot gets both: Codex keeps its own entry and skips any other server with that name.

What that agent already brings is on its own page, not the bot's. Open Extensions → Agents, choose the agent, and look under Sign in. Skills, Connectors, Rules and Plugins each show a count and open to the list. The section is left out when the agent has none of them. Those rows are read only. In the terminal, the same lists open from What the agent brings on the agent's sign-in sheet.

Skills

A skill is a folder with a SKILL.md file: a name, a line saying when to use it, and the instructions. A bot with a skill on is told its name, its description and where its instructions are, so it can read them when a task calls for it.

  • Install one: the Skills section of Extensions lists Anthropic's published skills. Press Add. Click any skill to read its instructions first.
  • Write your own: in the Skills section, choose Your own skill, above the list, and fill in Name, When to use it and Instructions.
  • Let a bot write one: ask a bot to save how it did something as a skill. A bot may also save one on its own after working out a task you're likely to want again. The chat shows "Saved the skill …" or "Updated the skill …". A skill a bot writes is on for that bot only; turn it on for other bots in their settings. A bot can change only the skills it wrote, and you can remove any skill from your library.

Use a skill with a slash command

Type / at the start of a message to see that bot's commands: built-in ones first (such as /fast for Fast mode, when the bot's agent offers it), then the skills the bot has on, then the commands of the bot's agent, marked with its logo (Commands). Picking a skill fills in /<skill id>; write the rest of your message and send it. The bot is told to read that skill first.

  • Desktop app: the list floats above the message box. Use ↑ and ↓, then Tab or Return. Escape hides it.
  • iPhone: the list sits above the message box. Tap a command.
  • Terminal UI: the list appears over the transcript. Use ↑ and ↓, then Tab or ↵.

Any other /command goes to the agent as you typed it, so your agent's own commands keep working. In a group chat, the list shows every member's skills, and a skill message reaches each member that has it on.

Automatic updates

Connectors installed from the registry as packages follow their upstream releases. Codync checks the registry from time to time and again just before the connector starts. When a newer version is published, it switches to it: a running connector keeps going, and the new version applies the next time it starts. If the check fails, the installed version stays, and Codync never moves a connector to an older version.

Custom connectors are never updated automatically.