Developing with AI Agents

You don't have to choose between "using the dashboard" and "writing code." Fiber AI ships first-class agent integrations, so your AI assistant can search companies, reveal contacts, and build audiences for you — inside the editor you already use.

Prerequisites

1. Get an API key at fiber.ai/app/api (starts with sk_live_), then export it. Details: API keys.

   export FIBER_API_KEY=sk_live_...

   Add it to your shell profile so it persists.

2. Have credits. Most operations consume credits — check your balance at fiber.ai/app/subscription and see Billing & credits.

Plugins (recommended)

The public plugin is fiber-ai/fiber-ai-plugin. One install wires three MCP servers (V2, V3, Core), 16 skills, and agent rules. Export FIBER_API_KEY first (V3 uses OAuth and can skip the key).

Claude Code:

/plugin marketplace add fiber-ai/fiber-ai-plugin
/plugin install fiber@fiber-tools

Already in a Claude Code chat? Send:

please install https://github.com/fiber-ai/fiber-ai-plugin

Gemini CLI:

gemini extensions install https://github.com/fiber-ai/fiber-ai-plugin

Cursor: paste an MCP deeplink into the browser (Cursor must be installed), or copy cursor/mcp.json from the plugin repo to .cursor/mcp.json. Full URLs: MCP setup by client.

OpenCode: tell the agent to fetch https://raw.githubusercontent.com/fiber-ai/fiber-ai-plugin/main/.opencode/INSTALL.md

Then try /fiber:help, or one of the workflow skills:

The rest of the 16 (quickstart, find-and-enrich-by-role, enrich-linkedin-csv, expand-from-email-list, enrich-github-handles, build-recruiting-audience, track-signals, benchmark-vs-competitor) are slash commands with the same /fiber: prefix. They ship with the plugin.

Team sharing

Commit an MCP config so the whole team gets Fiber automatically. Include all three servers; put the key in headers for V2 and Core, never in git:

{
  "mcpServers": {
    "fiber-ai-v2": {
      "type": "http",
      "url": "https://mcp.fiber.ai/mcp/v2",
      "headers": { "x-api-key": "${env:FIBER_API_KEY}" }
    },
    "fiber-ai-v3": {
      "type": "http",
      "url": "https://mcp.fiber.ai/mcp/v3"
    },
    "fiber-ai-core": {
      "type": "http",
      "url": "https://mcp.fiber.ai/mcp",
      "headers": { "x-api-key": "${env:FIBER_API_KEY}" }
    }
  }
}

Use .mcp.json at project root for Claude Code, .cursor/mcp.json for Cursor, .vscode/mcp.json for VS Code.

Individual MCP setup

If no plugin exists for your editor, add the MCP servers by hand — full per-editor instructions live on MCP for AI agents:

# Claude Code (V2 needs the key header; V3 is OAuth)
claude mcp add --transport http fiber-ai-v2 https://mcp.fiber.ai/mcp/v2 \
  --header "x-api-key: $FIBER_API_KEY"
claude mcp add --transport http fiber-ai-v3 https://mcp.fiber.ai/mcp/v3
claude mcp add --transport http fiber-ai-core https://mcp.fiber.ai/mcp \
  --header "x-api-key: $FIBER_API_KEY"

# Codex CLI — V2/Core need a bearer env var or Codex tries OAuth on V2
export FIBER_API_KEY=sk_live_...
codex mcp add fiber-ai-v2 --url https://mcp.fiber.ai/mcp/v2 \
  --bearer-token-env-var FIBER_API_KEY
codex mcp add fiber-ai-v3 --url https://mcp.fiber.ai/mcp/v3
codex mcp add fiber-ai-core --url https://mcp.fiber.ai/mcp \
  --bearer-token-env-var FIBER_API_KEY

OpenCode and VS Code use the same URLs in opencode.json / .vscode/mcp.json — copy the exact shapes from the MCP page.

Skills (from the plugin)

Install skills from fiber-ai/fiber-ai-plugin via skills.sh when you are not installing the whole plugin. These are on the skills.sh index today:

npx skills add fiber-ai/fiber-ai-plugin --skill search
npx skills add fiber-ai/fiber-ai-plugin --skill enrich
npx skills add fiber-ai/fiber-ai-plugin --skill audience
npx skills add fiber-ai/fiber-ai-plugin --skill setup
npx skills add fiber-ai/fiber-ai-plugin --skill help
npx skills add fiber-ai/fiber-ai-plugin --skill sdk-ts
npx skills add fiber-ai/fiber-ai-plugin --skill sdk-py

The other nine skills (including quickstart and the playbook workflows) live in the same plugin repo and install with the full plugin. Each skill teaches authentication, parameters, workflow order, and errors for that job so the agent does not invent routes.

SDK docs via Context7

When the agent is writing SDK code, add Context7 so it can pull current @fiberai/sdk / fiberai signatures and the @fiberai/sdk/zod schemas instead of guessing from training data:

claude mcp add --transport http context7 https://mcp.context7.com/mcp

Indexed libraries: TypeScript SDK and Python SDK.

Use Context7 for library docs; use Fiber MCP or the plugin for calling Fiber. Pair them: Context7 to write the call, Fiber MCP to run a lookup while you iterate. The operation index is still api.fiber.ai/ai-docs/index.md. See SDKs for the Zod parse → call pattern.

MCP vs SDKs: when to use which

Use both: prototype a search interactively over MCP, then ship it with the SDK.

Example prompts

Once connected, try:

• "Find AI companies in San Francisco with 50–200 employees"
• "Get the work email for linkedin.com/in/johndoe"
• "Build a list of 100 AI startups in the US and enrich their founders' emails"
• "Write a TypeScript script that searches companies and saves results to a database. Import Zod schemas from @fiberai/sdk/zod."
• "How many credits do I have left?"

Rate limits

Busy periods return HTTP 429 — wait briefly and retry. In production, add exponential backoff, and prefer the audience workflow for bulk work instead of many individual calls. Details in the API docs.