> ## Documentation Index
> Fetch the complete documentation index at: https://docs.runrally.io/llms.txt
> Use this file to discover all available pages before exploring further.

# Quickstart

> Create an agent key and connect Claude Code, Codex, Cursor, OpenClaw or any MCP client to Rally.

<Steps>
  <Step title="Create an agent key">
    In Rally, open **Settings → Agents & MCP** and click **Create key**.

    * **Name** it after where it will run, for example `Claude Code on Sam's laptop`.
    * **Role** decides what the agent may do. Start with **Program admin** unless the agent needs to move money or manage members.
    * Leave **Autonomous** off. The agent will then ask a person before high-risk actions.

    Copy the key when it is shown. It starts with `co_` and is only shown once.
  </Step>

  <Step title="Connect your client">
    The same page shows a ready-made snippet for each client with your URL and key filled in. These are the same snippets:

    <Tabs>
      <Tab title="Claude Code">
        ```bash theme={null}
        claude mcp add --transport http rally https://app.runrally.io/mcp \
          --header "Authorization: Bearer co_your_key"
        ```
      </Tab>

      <Tab title="Codex">
        ```toml ~/.codex/config.toml theme={null}
        [mcp_servers.rally]
        url = "https://app.runrally.io/mcp"
        bearer_token_env_var = "RALLY_AGENT_KEY"
        ```

        Then `export RALLY_AGENT_KEY=co_your_key` in the shell that runs Codex.
      </Tab>

      <Tab title="Cursor">
        Cursor limits how many tools a server can offer, so connect to the `core` profile (about 45 tools) rather than everything. See [Tools and scopes](/mcp/tools).

        ```json .cursor/mcp.json theme={null}
        {
          "mcpServers": {
            "rally": {
              "type": "streamable-http",
              "url": "https://app.runrally.io/mcp?profile=core",
              "headers": { "Authorization": "Bearer co_your_key" }
            }
          }
        }
        ```
      </Tab>

      <Tab title="OpenClaw">
        ```bash theme={null}
        openclaw mcp add rally --url https://app.runrally.io/mcp --transport streamable-http
        ```

        Or in `~/.openclaw/openclaw.json`:

        ```json theme={null}
        {
          "mcp": {
            "servers": {
              "rally": {
                "url": "https://app.runrally.io/mcp",
                "transport": "streamable-http",
                "headers": { "Authorization": "Bearer co_your_key" }
              }
            }
          }
        }
        ```
      </Tab>

      <Tab title="Other clients">
        Any client that speaks MCP over Streamable HTTP works.

        | Setting | Value |
        | - | - |
        | Transport | Streamable HTTP |
        | URL | `https://app.runrally.io/mcp` |
        | Header | `Authorization: Bearer co_your_key` |

        Clients that support OAuth can skip the key: add the URL and sign in with your Rally account when the client asks.
      </Tab>
    </Tabs>
  </Step>

  <Step title="Say hello">
    Ask the agent:

    > Who am I connected as in Rally, and what is left to set up?

    It calls `rally_whoami` and `rally_setup_status` and reports the workspace, its role, your plan and the open setup steps. From there, ask for what you want in plain words: "list my programs", "draft a brief for a TikTok launch video", "who is behind schedule this week".
  </Step>
</Steps>

<Tip>
  Agents work best when they read Rally's runbook first. Ask it to read `https://app.runrally.io/mcp/skill.md` (or the `rally://docs/mcp` resource the server offers) before its first task.
</Tip>

## Next

<CardGroup cols={2}>
  <Card title="Tools and scopes" icon="wrench" href="/mcp/tools">
    Connect a smaller set of tools, and learn the helper tools.
  </Card>

  <Card title="Approvals and roles" icon="shield-check" href="/mcp/approvals">
    What waits for a person, and how to approve it.
  </Card>
</CardGroup>


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.