Integrating MCP V2
Firefly MCP Server — Setup & Integration Guide
AWS Bedrock AgentCore MCP Gateway
Overview
What Is It?
The Firefly MCP Server is an AWS Bedrock AgentCore MCP Gateway service that enables AI agents — such as Claude Code, Cursor, and Claude Desktop — to access Firefly backend APIs in a secure, standardized way.
AI agents authenticate using a Firefly Access Key and Secret Key. A Lambda interceptor handles credential exchange, converting these keys into a short-lived JWT that is transparently attached to every forwarded request.
The gateway is automatically synced with the latest Firefly API specs.
How It Works
AI Agent → AgentCore MCP Gateway → Lambda Interceptor (JWT exchange) → Firefly Backend API
Agent sends
x-access-keyandx-secret-keyheaders.Lambda exchanges credentials for JWT (cached until expiry).
JWT is added as
Authorization: Bearerheader.Request is forwarded to the backend API.
Step 1 — Create API Keys
Before configuring any client, you need an API Key pair. You can use an existing key pair or generate a new one.
Use Existing Key Pair
Enter your Access Key and Secret Key in the provided fields. If you already have a Firefly API key pair (for example, from Settings → Access Management), you can use it here.
Generate New Key Pair
Generate a new API Key pair directly from the MCP integration page:
In the Key Type dropdown, choose one of:
User API Key — Creates a new API key associated with your user account.
Service Account API Key — Creates a new API key associated with a service account. Select the desired service account from the list, then use New API Key to create the key for that account.
Click New API Key (or complete the flow for the selected key type). Your ACCESS KEY and SECRET KEY will be displayed after the keypair is created.
Important: When generating a new key, save your Secret Key immediately. Once you close the dialog, the Secret Key cannot be retrieved again. Store it securely in a password manager or secrets vault.
MCP configuration
No matter which method you choose—existing key pair or newly generated (User or Service Account)—the MCP configuration snippet on the page is automatically updated with your Access Key and Secret Key. You can copy that snippet into Cursor, Claude Code, or Claude Desktop as described in the steps below.
Step 2 — Choose Your MCP Endpoint
The Firefly MCP Server is available in two regions. Use the endpoint that matches the region your Firefly app is hosted in:
US —
https://mcpv2.firefly.ai/mcp(app atapp.firefly.ai)EU —
https://mcpv2.eu.firefly.ai/mcp(app atapp.eu.firefly.ai)
Note: The
X-access-Key/X-secret-Keyheaders are identical for both regions. In the client setup steps below, replace<YOUR_MCP_ENDPOINT>with the endpoint for your region.
Cursor Setup
Open mcp.json
Locate and open your mcp.json configuration file.
Add the MCP Server Configuration
Insert the following JSON block, replacing the placeholder values with your credentials:
Restart Cursor
Save the file and restart Cursor.
Verify the Installation
Go to Settings → Tools and MCP. Under Installed MCP Servers you should see firefly-mcp.
Claude Code Setup
Add the MCP Server
Run the following command in a terminal, replacing the placeholder values with your credentials:
Verify the Installation
firefly-mcp should appear in the output.
Codex Setup
The OpenAI Codex CLI connects to the Firefly MCP Server over streamable HTTP. You can add it by editing config.toml directly or by using the codex mcp add command.
Note: Remote (HTTP) MCP servers require a recent version of Codex CLI. Check your version with
codex --versionand upgrade if needed (e.g.npm install -g @openai/codex@latest).
Option A — Edit config.toml (recommended)
Open config.toml
Locate and open ~/.codex/config.toml in a text editor. Create the file if it does not already exist.
Add the MCP Server Configuration
Add the following block, replacing the placeholder values with your credentials:
To keep your credentials out of the config file, set them as environment variables and reference them with env_http_headers instead. Codex resolves each header value from the named environment variable at runtime:
Option B — Use the Codex CLI
If your Codex version does not yet support remote HTTP servers, you can connect through the mcp-remote bridge, which runs as a local stdio server. This mirrors the Claude Desktop setup below and requires npx (Node.js).
Run the following command in a terminal, replacing the placeholder values with your credentials:
Verify the Installation
Start Codex and run /mcp in the TUI to list active MCP servers. firefly-mcp should appear. You can also confirm the entry was written to ~/.codex/config.toml.
Claude Desktop Setup
Prerequisites
npx must be installed on your machine. npx is included with Node.js. We recommend installing the latest version of Node.js. Verify by running
npx --versionin a terminal.
Open Claude Desktop Settings
Press your user avatar in the bottom left corner and select Settings.
Navigate to the Developers Tab
Go to the Developers tab and click Edit Config. A new window will open the Claude configuration directory.
Edit the Configuration File
Open claude_desktop_config.json with a text editor and add the following:
Restart Claude Desktop
Save the changes and restart Claude Desktop.
Verify the Installation
Go to Settings → Developers tab. Under Local MCP servers you should see firefly-mcp.
Limitations
Note: The tools presented in any connected client may not reflect the full set of available tools exposed by the Firefly MCP Server. You'll see 63 tools.
Last updated
Was this helpful?