The Model Context Protocol (MCP) server provides a standardized interface that allows any compatible AI agent to access Fiber AI's data and tools — search companies, enrich contacts, reveal emails, and more — directly from your editor.
For AI agents
If you are a coding agent and need to pick between MCP and the REST API,
start with the canonical agent-facing docs on the API:
Search companies, enrich contacts, and reveal emails and phones from your AI agent.
Rule of thumb: use MCP when you're acting inside an IDE / chat and
each operation is a tool call; use the REST API (via
@fiberai/sdk or
fiberai) when you're building
an autonomous script or a production pipeline.
Servers
Fiber AI offers three remote MCP servers:
Server
URL
Auth
Best For
V2
https://mcp.fiber.ai/mcp/v2
API key
Auto-generated direct tools for the top ~10 priority operations (companySearch_tool, peopleSearch_tool, etc.)
V3
https://mcp.fiber.ai/mcp/v3
OAuth (SSO)
Direct tools for every public operation with compact descriptions; the model can call or expand any tool on demand
Core
https://mcp.fiber.ai/mcp
API key
5 meta-tools that discover and call any of 100+ API endpoints (search_endpoints, list_tag_packs, list_all_endpoints, get_endpoint_details_full, call_operation)
Which one should I use?V2 is the fastest path for the most common flows with an API key. V3 is the most ergonomic for broad agent coverage — it exposes every public operation as a direct tool with compact descriptions, authenticated via browser-based SSO login instead of a pasted key. Core keeps the tool count tiny (5) and lets the agent discover endpoints at runtime; it uses the same API-key auth as V2. You can register more than one; the server names (fiber-ai-v2, fiber-ai-v3, fiber-ai-core) stay distinct.
Want more than MCP? Use the Fiber AI plugin
The Fiber AI plugin bundles these MCP servers with 16 skills (search, enrich, audiences, recruiting and SDR playbooks) and 7 persona subagents (ai-sdr, ai-recruiter, gtm-strategist, and more) for Claude Code, Cursor, Gemini CLI, OpenCode, Codex, and VS Code. See its install guide. This repo covers the MCP servers only.
Setup
Get an API key at fiber.ai/app/api for V2 and Core. The examples read it from a FIBER_API_KEY environment variable. V3 needs no key: your client opens a browser for SSO login on first connect.
Connect directly with the config for your client below, or through Smithery.
Replace cursor with your client: claude-code, claude, vscode, codex, gemini-cli, windsurf, opencode, cline, goose, and more (npx -y @smithery/cli@latest mcp add --help lists them all). Use fiberai/fiber-core for Core. On first connect your client opens Smithery in the browser to sign in and set your Fiber API key.
To use it as a Smithery connection instead (for agents that call tools through the Smithery CLI):
Add -s user to install for every project instead of the current one. Run gemini mcp list to verify, and /mcp auth fiber-ai-v3 inside a session to sign in to V3.
OpenCode
Add to opencode.json in your project (or ~/.config/opencode/opencode.json):
Auth: header x-api-key: <your key> (or Authorization: Bearer <your key>) for V2 and Core; OAuth for V3
For stdio-only clients: npx -y mcp-remote https://mcp.fiber.ai/mcp/v2 --header x-api-key:$FIBER_API_KEY
Available Tools
V2 Server (/mcp/v2)
Direct API tools — each Fiber AI endpoint is exposed as an individual tool:
Tool
Description
companySearch_tool
Search for companies by industry, location, size, funding, etc.
peopleSearch_tool
Search for people by title, seniority, department, etc.
profileLiveEnrich_tool
Live-fetch a LinkedIn profile
companyLiveEnrich_tool
Live-fetch a LinkedIn company
getOrgCredits_tool
Check your credit balance
getRateLimits_tool
Check your organization's rate limits
getAutoTopupSettings_tool
Get auto top-up settings
getAllApiKeys_tool
List your API keys
cardsAttach_tool
Attach a card to a trial account
getQuickstartScript_tool
Get the quickstart install script
V3 Server (/mcp/v3)
One direct tool per public operation (100+), with compact descriptions the model can expand on demand.
Core Server (/mcp)
Meta tools for dynamic endpoint discovery:
Tool
Description
search_endpoints
Search for API endpoints by keyword
list_tag_packs
List curated groups of endpoints to narrow the catalog
list_all_endpoints
List all available API endpoints
get_endpoint_details_full
Get full schema details for an endpoint
call_operation
Call any API endpoint by its operation ID
Authentication
V2 and Core require a Fiber AI API key. V3 uses OAuth (Clerk SSO) - the client opens a browser window for login and reuses the session token, no key configuration needed. Get an API key at fiber.ai/app/api.
You can supply your key in any of these three ways - pick whichever your client makes easiest:
Option A - via chat (zero config)
The agent passes the key in the request body as apiKey. Simplest path: paste your key into the agent once and it'll keep using it.
You: Use sk_live_... as my Fiber API key.
Agent: [calls companySearch_tool with { "apiKey": "sk_live_...", "searchParams": {...} }]
Caveat: the key ends up in chat history. Fine for personal use, not ideal for shared sessions.
Option B - via x-api-key header (recommended for IDEs)
Configure the header once in your MCP client config; the agent never sees the key.
Most clients accept a headers field on each MCP server and can read the key from an environment variable instead of hard-coding it. The field name and variable syntax differ per client; see the Setup section for yours.
Option C - via Authorization: Bearer header
Same shape as Option B, but using the standard bearer-token header. Useful if your MCP client only supports Authorization-style auth.
When more than one is present, the server resolves in this order: body/query apiKey -> x-api-key header -> Authorization: Bearer. The first non-empty value wins.
V3 (OAuth)
V3 ignores all of the above. On first connect, the client opens https://app.fiber.ai for browser-based SSO login; the resulting session token is reused on subsequent calls. No env vars, no headers, no config to write.
Example Usage
Once connected, ask your AI agent:
"Search for SaaS companies in New York with 50-200 employees"
"Find the CEO of and get their work email"
"How many credits do I have left?"
"Enrich this LinkedIn profile: linkedin.com/in/..."
SDKs
For building applications programmatically (not via MCP):
Connection not working?
Ensure your editor supports HTTP (Streamable HTTP) MCP transport. If it only supports stdio, use the npx mcp-remote wrapper shown under Others.
Getting authentication errors?
Make sure you're supplying a valid key via one of the three supported paths (body apiKey, x-api-key header, or Authorization: Bearer). See Authentication above. Get a key at fiber.ai/app/api.
Should I put the key in chat or in headers?
For personal sessions, chat is fine. For shared sessions, team-config files committed to git, or anywhere chat history might leak, configure the x-api-key header at the MCP client layer so the agent never sees the raw key.
Can I use more than one server at the same time?
Yes. You can register V2, V3, and Core side by side; their names stay distinct.