AI connections (MCP)

Overview

AI connections let you plug your favorite AI assistant — Claude, ChatGPT, Claude Code or any client supporting the Model Context Protocol (MCP) — directly into your Obside account.

Once connected, your assistant can work with your real Obside data and act on your behalf:

• Read your portfolios — positions, performance, exposure, income forecast, related news, calendar.
• Look up markets — search assets, get prices and price history.
• Manage your agents — list, create, update, activate or stop agents, organize them in folders.
• Import TradingView Pine Scripts: Obside runs them as-is, and your assistant builds agents on them.
• Run backtests and read their results.
• Place or cancel orders on your connected accounts (same regional and plan rules as in the app).
• Check your credit balance.

---

How to connect

Everything happens in Settings → AI connections, where the MCP server URL is displayed with step-by-step guides for the most popular clients.

Clients with sign-in (Claude, ChatGPT…)

Most modern AI clients connect with a simple sign-in — no token needed:

1. In your AI client, add a custom connector and paste Obside's MCP server URL.
2. Leave any "Client ID" or "Client Secret" fields empty.
3. Sign in to Obside when prompted to authorize the connection.

For example, in Claude: Settings → Connectors → Add custom connector. In ChatGPT: enable Developer mode, then Settings → Connectors → Create.

Clients using a token (Claude Code, Codex…)

Some clients authenticate with a bearer token instead of a sign-in:

1. In Settings → AI connections, generate a token and copy it.
2. Add the Obside MCP server in your client with the token as the Authorization header — the settings page shows the exact ready-to-paste command per client.

For Claude Code, add it with the token as a header:

claude mcp add --transport http --scope user obside https://api.obside.com/mcp \
  --header "Authorization: Bearer YOUR_TOKEN"

For Codex, declare the server in ~/.codex/config.toml and keep the token in an environment variable (Codex does not expand placeholders inside the file):

[mcp_servers.obside]
url = "https://api.obside.com/mcp"
bearer_token_env_var = "OBSIDE_TOKEN"

You can name your tokens (one per client is a good habit), see when each was last used, and revoke any of them at any time.

Writing agent code from your assistant

Claude Code and Codex can go further than reading and managing: they can write the code behind an agent and hand it to Obside. See Build with Claude Code and Codex for the full workflow and the reusable skill, and the Agent code (STSE) reference for the runtime itself.

The connection itself also serves that material, so there is nothing to install. The skill and the developer reference pages are exposed as MCP resources (the skill is obside://skills/obside-stse), and a build_agent prompt starts a conversation with the skill already loaded and your idea attached. In Claude Code, the prompt shows up as the slash command /mcp__obside__build_agent and the resources can be attached with an @ mention. In a chat app such as claude.ai or Claude Desktop, attach a resource yourself or run the prompt, since the assistant does not load a resource on its own. Clients that only list tools simply ignore them.