A plugin that gives an AI coding agent Deriv API knowledge. Installing it as a native plugin for Cursor, Claude Code, and Codex adds the hosted MCP server plus bundled Agent Skills. Agents that only support the Agent Skills standard can install the skills separately.
/plugin marketplace add deriv-com/deriv-api-plugin
/plugin install deriv@deriv
The first command registers the public repository as a plugin marketplace; the second installs the plugin from it. No Node or Python is required. Installing the plugin approves its MCP server once, at install time; Claude Code may still prompt per tool call.
The plugin is listed on the Cursor Marketplace at https://cursor.com/marketplace/deriv. Install it from there:
/add-plugin deriv
As a secondary option — to try a surface that is not yet published to the marketplace — clone the public repository into Cursor's local plugin directory:
git clone https://github.com/deriv-com/deriv-api-plugin ~/.cursor/plugins/local/deriv
Then reload the Cursor window so the plugin is picked up.
codex plugin marketplace add deriv-com/deriv-api-plugin
That registers this repository as a Codex marketplace. Install deriv from the
Plugins Directory, then restart Codex so it picks up the hosted MCP server. To
refresh an existing install after a listing change, run codex plugin marketplace upgrade.
If Codex shows an authentication step during install, that is Codex's
generic MCP install flow — the plugin itself needs no Deriv sign-in
and asks for no API token.
In Claude Code, run /mcp and confirm deriv shows as connected, with its
tools listed. In Cursor, open the MCP settings and confirm the same. If the
hosted endpoint is unreachable, the bundled skills still load; follow those
instead of inventing field lists. Do not invent field lists in either case.
This path copies skills/ only and does not add the hosted MCP.
If your agent cannot install a plugin, add the skills with:
npx skills add https://github.com/deriv-com/deriv-api-plugin
That command is the npx skills CLI at https://skills.sh.
Or copy the folders under skills/ into the personal skills directory for your
agent:
| Agent | Skill directory |
|---|---|
| Claude Code | ~/.claude/skills/ |
| Cursor | ~/.cursor/skills/ |
| OpenCode | ~/.config/opencode/skills/ |
| Codex | ~/.codex/skills/ |
| Pi | ~/.pi/agent/skills/ |
If the plugin is already installed on that host, do not also copy skills or run
npx skills. Doing both can load the same skills twice.
| Skill | Useful for |
|---|---|
deriv-trading-app |
Build or extend an authenticated trading application |
deriv-auth |
Authentication, OAuth, tokens, accounts, and sessions |
deriv-market-data |
Market discovery and live prices |
deriv-trade-types |
Map user-facing trade names to contract families |
deriv-trade-lifecycle |
Proposal, buy, monitor, sell, and settlement |
deriv-lightweight-charts |
Add the default slim price chart with barriers and trade markers |
deriv-smartcharts |
Add the terminal-grade chart with indicators and drawing tools |
deriv-llms |
Bundled docs snapshot when hosted MCP is unavailable |
| Server key | URL | Purpose |
|---|---|---|
deriv |
https://mcp-api-v2.deriv.com/mcp |
Live schemas, field facts, payload validation, and task guides |
The server is declared twice at the plugin root, for the same key at the same
URL: .mcp.json (type: "http") is what Cursor and Claude Code read, and
mcp.json (type: "streamable-http", under the Agent Plugins v1 schema) is
what a host following that standard reads. Codex reads the root mcp.json.
The Codex compatibility manifest's mcpServers field names ./.mcp.json, but
this is a portable package, and under
OpenAI's package rules a portable
package's root mcp.json is what OpenAI and Codex discover; that field cannot
replace it. That matches what we measured on 2026-10-01 with Codex CLI 0.159.2.
Both files name the same server at the same URL.
Once installed, the plugin's MCP tools, the Cursor rule, and the bundled skills
are available to the agent automatically. No credentials and no configuration
are required — the plugin requires no Deriv account credentials. The hosted
server reads public Deriv API documentation and schemas to answer questions and
validate payloads; it does not sign in, store keys, or ask you to set anything
up. Tool-call arguments, including a validate_payload body, reach the hosted
server, so keep API tokens and other secrets out of tool-call arguments.
When you build an integration that itself calls the Deriv API, your own
application supplies its own credentials.
CONTRIBUTING.md— how to file issues and open pull requests.SECURITY.md— network scope, credential handling, and how to report a vulnerability.PRIVACY.md— what data the plugin processes, and the Deriv Privacy Policy that applies.LICENSE— MIT.