The best way to control Claude Code, Codex and more from any computer or phone
Mirrored from https://github.com/Camo67/t3code.git
Synced Nov 21, 03:37 UTC
| 1 | # Outside agents |
| 2 | |
| 3 | Agents that T3 Code did not start can work with an environment through its MCP |
| 4 | server. They can read projects and threads, start and message threads, and |
| 5 | check which providers and models are available. This covers Claude Code or |
| 6 | Codex in your own terminal, ChatGPT, and bots that support MCP. Each agent signs |
| 7 | in once, and you choose what it may do. |
| 8 | |
| 9 | ## Add T3 Code to an assistant app |
| 10 | |
| 11 | Open the command palette (`Cmd/Ctrl+K`) and choose **Use T3 Code in another |
| 12 | app (MCP)**, or use the button in **Settings → Sharing & access**. Pick the |
| 13 | app and follow the step for that app. Assistants like |
| 14 | Poke and Claude run in the cloud, so the wizard first asks you to sign in to T3 |
| 15 | Connect and turn it on for the environment. To reach your other machines through |
| 16 | the same connection, turn on **Linked environments** in Labs and |
| 17 | [link them](./remote-access.md#link-another-environment). |
| 18 | |
| 19 | When the app opens the sign-in page, choose what it may do and approve with your |
| 20 | T3 Connect account. On an older environment, the wizard asks for the level and |
| 21 | gives you a code to paste instead. A code works once, for 5 minutes, and |
| 22 | approves only that level. |
| 23 | |
| 24 | For an app that is not listed, choose **Something else…**, add a remote MCP |
| 25 | server with OAuth in that app, and paste the URL. Apps that require their own |
| 26 | client ID cannot connect. |
| 27 | |
| 28 | ## Get the MCP URL |
| 29 | |
| 30 | In **Settings → Environments**, open a saved environment's menu and choose |
| 31 | **Copy MCP URL**. The URL is the environment's address followed by `/mcp`, for |
| 32 | example: |
| 33 | |
| 34 | ```text |
| 35 | https://<environment-address>/mcp |
| 36 | ``` |
| 37 | |
| 38 | When T3 Connect serves the environment, the copied URL is its T3 Connect |
| 39 | address, which agents can reach from anywhere. Otherwise it uses the route this |
| 40 | device is connected over, so an agent on the same device can reach it. |
| 41 | |
| 42 | - **An agent on your own computers** can use `localhost` on the host, T3 |
| 43 | Connect, or any `https` address, such as |
| 44 | [Tailscale HTTPS](./remote-access.md#tailscale-https). Signing in sends the |
| 45 | agent's credentials to the environment, and agents only do that over `https` |
| 46 | or to `localhost`. A plain-`http` LAN or Tailscale address, like |
| 47 | `http://100.x.y.z:3773`, lets an agent find the environment but not finish |
| 48 | signing in. |
| 49 | - **A hosted agent**, such as ChatGPT or a bot running in the cloud, must reach |
| 50 | the environment from the internet. Use the T3 Connect address. A Tailscale |
| 51 | address only works from your own tailnet. |
| 52 | |
| 53 | ## Approve a sign-in |
| 54 | |
| 55 | The first time an agent connects, it opens a sign-in page. When the agent |
| 56 | reaches the environment through T3 Connect, the page opens on the T3 Code web |
| 57 | app: sign in with the T3 Connect account the environment is linked to and |
| 58 | approve there, without a code. Otherwise the page opens on the environment. |
| 59 | Either page names the agent and where its access goes: |
| 60 | |
| 61 | - For an agent on your computer, approval returns to a `localhost` address on |
| 62 | the computer that opened the page. |
| 63 | - For a hosted agent, approval goes to that service, such as `chatgpt.com`. |
| 64 | Anyone who runs that service gets the access. |
| 65 | |
| 66 | Only approve a sign-in you just started. On the environment's own page, enter a |
| 67 | code from **Use T3 Code in another app (MCP)**, **Settings → Sharing & access**, |
| 68 | or `t3 auth pairing create` on the host. A browser already signed in to the |
| 69 | environment with permission to manage access can approve without a code. On the |
| 70 | web app, **Use a sign-in code instead** opens the environment's own page. |
| 71 | |
| 72 | A code from **Use T3 Code in another app (MCP)** already carries the level you |
| 73 | picked there. With another code, choose what the agent may do here: **Auto** is |
| 74 | selected by default, or the closest lower mode allowed by your organization’s |
| 75 | policy. Approving without a code starts at **Read only**: |
| 76 | |
| 77 | - **Read only** lets the agent read projects and threads and see |
| 78 | the available providers and models. It cannot change anything. |
| 79 | - **Supervised** through **Full access** also let it start, message and stop |
| 80 | threads in every project. A thread it starts or steers cannot run with more |
| 81 | permissions than the mode you chose. |
| 82 | |
| 83 | ## What outside agents can call |
| 84 | |
| 85 | Some tools only work for an agent running inside a T3 thread, because they act |
| 86 | as that thread: subagents, secret requests, rendered HTML, the browser preview, |
| 87 | devices, and worktree handoff. They still appear in an outside agent's tool |
| 88 | list, but calling one returns an error saying it needs an agent running inside |
| 89 | T3 Code. |
| 90 | |
| 91 | - To hand off work, start a thread with `t3_thread_launch`, then follow it with |
| 92 | `t3_thread_wait` or `t3_thread_read`, instead of `create_threads` or |
| 93 | `delegate_task`. |
| 94 | - Tools that default to the caller's own project or thread need an explicit |
| 95 | `projectId` or `threadId`. |
| 96 | - To list threads across projects, call `t3_thread_list` with |
| 97 | `allProjects: true`, or use `t3_thread_search`. |
| 98 | |
| 99 | ## Claude Code |
| 100 | |
| 101 | ```sh |
| 102 | claude mcp add --transport http t3 https://<environment-address>/mcp |
| 103 | claude mcp login t3 |
| 104 | ``` |
| 105 | |
| 106 | `claude mcp login` opens the sign-in page in your browser. On a machine |
| 107 | without a browser, add `--no-browser`, open the printed URL elsewhere, and |
| 108 | paste the final URL back when asked. `claude mcp list` shows `t3` as connected |
| 109 | once you approve. |
| 110 | |
| 111 | ## Codex |
| 112 | |
| 113 | ```sh |
| 114 | codex mcp add t3 --url https://<environment-address>/mcp |
| 115 | codex mcp login t3 |
| 116 | ``` |
| 117 | |
| 118 | `codex mcp login --no-browser` prints the sign-in URL instead of opening it. |
| 119 | `codex mcp list` shows `t3` with OAuth once you approve. |
| 120 | |
| 121 | ## ChatGPT |
| 122 | |
| 123 | Add the environment as a custom MCP app in ChatGPT's apps settings: |
| 124 | |
| 125 | 1. Create a new app or plugin with a custom MCP server. |
| 126 | 2. Set the server URL to `https://<environment-address>/mcp`, using the T3 |
| 127 | Connect address. |
| 128 | 3. Choose **OAuth** for authentication. ChatGPT discovers the rest; you do not |
| 129 | need a client ID or secret. |
| 130 | 4. Create the app, then approve the sign-in page that opens. |
| 131 | |
| 132 | ## Other agents and bots |
| 133 | |
| 134 | Any agent that supports remote MCP servers over HTTP with OAuth can connect. Give |
| 135 | it the MCP URL and choose OAuth, and the agent finds the sign-in settings itself. |
| 136 | Agents without OAuth support cannot connect. |
| 137 | |
| 138 | ## Scripts on the host |
| 139 | |
| 140 | On the machine that runs the environment, `t3 thread` uses the same tools |
| 141 | without a sign-in: |
| 142 | |
| 143 | ```sh |
| 144 | id=$(t3 thread start . "Find why the tests are slow") |
| 145 | t3 thread send "$id" "Now fix it" --wait |
| 146 | t3 thread show "$id" |
| 147 | t3 thread list |
| 148 | ``` |
| 149 | |
| 150 | `.` means the project that contains the current folder, or a git worktree of |
| 151 | its repository; a project id also works. A new thread uses the project's |
| 152 | default model and permission mode, and runs in the project folder unless you |
| 153 | pass `--worktree <base-branch>`. `send` never interrupts: a busy thread gets |
| 154 | the message as its next turn. Pass `-` as the message to read it from stdin. |
| 155 | |
| 156 | To choose the model, pass `--provider <instance-id> --model <slug>`. Add |
| 157 | `--option id=value` for each model option, such as `--option effort=high`. |
| 158 | Option ids and values depend on the model; `true` and `false` set on/off |
| 159 | options. |
| 160 | |
| 161 | `--wait` prints the agent's reply. It exits 1 when the run fails or stops, and |
| 162 | 2 when the agent waits for an answer you give in the app. Add `--json` for a |
| 163 | result that always includes the thread id. |
| 164 | |
| 165 | ## Remove an agent |
| 166 | |
| 167 | Approved agents appear under **Settings → Sharing & access** like other clients. |
| 168 | Revoke one there to cut off its access immediately. A sign-in lasts 30 days; |
| 169 | after that the agent asks you to approve it again. |