MCP Outreach Tools: How to Configure Model Context Protocol for Sales

Building custom integrations for AI agents is a major challenge for modern sales engineering teams. Every time you connect a model to a new tool, you must write API endpoints, document JSON schemas, and program custom logic. This manual setup is slow and makes it difficult to upgrade to new models.

The solution to this integration bottleneck is the Model Context Protocol (MCP). MCP is an open-source standard that defines how client applications connect to server tools. Instead of writing custom code, servers provide machine-readable schemas, allowing agents to discover and run tools automatically.

Omentir adopts this standard by hosting a dedicated MCP server. By connecting your AI agent to this server, you give it access to your LinkedIn outreach stack, allowing it to manage campaigns directly from your workspace. Let's look at how to configure your MCP clients.

The important distinction is that MCP is not another automation shortcut. It is a clean contract between the agent and the sales system. The agent can see which tools exist, what each tool expects, and what the tool returned. That structure matters because outreach work has real consequences: the wrong lead group, the wrong campaign state, or the wrong reply can affect an actual prospect conversation.

For teams that prefer standard REST integrations, the platform also provides full OpenAPI-compliant endpoints under /api/agent/v1. However, using the native MCP server provides the most seamless experience for modern developer environments, as detailed in our guide to Omentir's agent features.

The Omentir Hosted MCP Server Architecture

The Omentir MCP server runs as a hosted endpoint. Unlike local MCP servers that require Node.js or Python packages on your computer, the hosted server manages credentials and API pathways in the cloud.

This hosted design makes configuration simple. You do not need to run terminal scripts or install external packages. You simply point your local client (like Cursor or Claude CLI) to the Omentir MCP URL and pass your workspace-scoped token as authorization.

In practice, the hosted server sits between your agent client and your Omentir workspace. The client asks for available tools, the server returns tool schemas, and the agent chooses a tool call with structured arguments. Omentir then applies the same workspace checks, campaign rules, and pacing safeguards used by the rest of the product.

This is especially useful for sales teams that want an AI assistant to help with operations without handing it unrestricted access. A founder can ask an agent to inspect campaign readiness, create a draft campaign, review conversations, or summarize lead groups. The agent gets the specific tools it needs, while billing, account credentials, and unrelated workspace settings stay outside the tool surface.

Step-by-Step Client Setup for Cursor, Claude, and ChatGPT

Setting up your client takes only a few minutes. The exact screen names can change as agent products evolve, but every setup needs the same three pieces: the Omentir MCP endpoint, your workspace token, and a supported transport method.

Start by creating a fresh token for this use case rather than reusing an old one. Name it after the client, such as "Claude desktop prospecting" or "Cursor campaign QA." That makes later cleanup simple. If you stop using a client, revoke that token without affecting other agent workflows.

Integrating with the Cursor IDE Agent

To connect Cursor to Omentir, open your settings panel and navigate to the MCP server settings. Add a new server and enter the following settings:

• Name: omentir
• Type: SSE
• URL: https://omentir.com/api/agent/v1/mcp?key=[your_omentir_agent_token]

Once added, Cursor will discover Omentir's tools automatically, allowing you to ask the IDE agent to build campaigns or check responses directly from your editor.

A good first prompt is deliberately boring: "Call omentir_get_context and summarize whether the workspace is ready for automation." If the agent can read workspace context, authentication is working. If it cannot, fix the token or client transport before asking it to create anything.

Integrating with the Claude Code CLI

To connect Claude Code, add Omentir as an HTTP MCP server using your client configuration or command-line setup. A JSON client configuration usually follows this shape:

{
  "mcpServers": {
    "omentir": {
      "url": "https://omentir.com/api/agent/v1/mcp",
      "headers": {
        "Authorization": "Bearer <your_omentir_agent_token>"
      }
    }
  }
}

Save the file and restart your CLI session to verify the connection.

For ChatGPT or clients that do not support custom headers, use the query-parameter fallback: https://omentir.com/api/agent/v1/mcp?key=<your_omentir_agent_token>. Treat that URL like a password. Do not paste it into shared docs, screenshots, or public issue trackers.

Deconstructing the Omentir MCP Tool Catalog

The Omentir MCP server provides 19 tools that cover the entire outbound workflow. These tools are grouped into five functional categories:

• Context & Profile: Tools like omentir_get_context and omentir_get_product_profile that read your target buyer parameters and feature details.
• Lead Discovery: Tools like omentir_create_agent and omentir_list_leads that search LinkedIn daily and score fit profiles.
• Campaign Builder: Tools like omentir_create_campaign that build, edit, and check message sequences.
• Lifecycle Management: Tools to pause, resume, or remove discovery agents and campaign states.
• Replies: Tools like omentir_list_conversations and omentir_reply_to_lead that monitor your inbox and draft responses.

Think of these categories as a workflow, not a menu. A sensible agent run begins with context, checks the product profile, inspects existing agents or campaigns, then decides whether to create something new. Jumping straight to campaign creation is usually a bad prompt because the agent has not yet learned the workspace constraints.

A strong operator prompt might be: "Read the workspace context and product profile. List active lead groups. If there is no group for early-stage SaaS founders selling to sales teams, propose a draft discovery agent and explain the targeting assumptions before creating it." That prompt keeps the agent anchored to evidence and asks for assumptions before action.

For detailed campaign configuration steps, read our guide to MCP B2B campaign design.

Workspace Security and Token Authorization Boundaries

Security is critical when delegating tasks to AI agents. Omentir protects your workspace by enforcing strict authorization boundaries on all agent tokens.

Each token is scoped to a single workspace. It cannot edit billing details, view other accounts, or access your raw LinkedIn credentials. You can revoke tokens anytime in Settings → AI Agents, disabling access immediately.

You should still treat agent access as production access. Use one token per client, rotate tokens when a teammate leaves, and keep the token out of prompts that might be saved in transcripts. If you are testing a new agent, begin with read-heavy tasks such as context checks, campaign review, or dry-run readiness checks before allowing it to create or resume campaigns.

The safest pattern is to separate planning from execution. Ask the agent to inspect context and propose changes first. Then ask it to apply the change only after you have reviewed the plan. This keeps the workflow fast while still preventing a model from quietly acting on a bad assumption.

This security structure allows you to connect external agents without risking account safety. For integration blueprints, read our guide on integrating autonomous sales agents.

Pacing Campaign Deliveries to Maintain Account Health

LinkedIn monitors both the volume and speed of connection requests and messages. If you send too many invites in a short period, your profile will be restricted.

To protect your account, configure campaigns around conservative daily quotas, natural sending windows, and reviewable drafts. Omentir manages these safety protocols automatically, coordinating outgoing messages through secure, human-paced queues.

MCP does not change that philosophy. An agent can help build and inspect the outbound system, but it should not turn LinkedIn into a bulk sender. The best use of agent tooling is to improve lead fit, message relevance, and response handling while keeping the actual delivery pace human.

Before any campaign moves from draft to active, run a dry check. Confirm the workspace has a connected LinkedIn account, the lead group has qualified prospects, the sequence has sensible copy, and the campaign state is intentional. If the dry run reports a blocker, fix the blocker rather than asking the agent to work around it.

SOP: The MCP Client Setup and Audit Checklist

Follow this simple SOP to configure and audit your client setups:

• Mint Token: Go to Settings → AI Agents to generate a workspace token.
• Select Client Type: Choose SSE (Server-Sent Events) for editor integrations (Cursor, VS Code) or JSON block for local configs (Claude Code).
• Check Connection: Run a basic context tool call (e.g., omentir_get_context) to verify authentication.
• Review Limits: Set campaign discovery agent quotas within daily safety boundaries.
• Monitor Logs: Check Omentir logs to confirm that outgoing messages are sent using natural pacing.

Add two checks for real-world usage. First, ask the agent to explain what it is about to do before it creates or resumes anything. Second, keep a short changelog of agent-created campaigns: who prompted the change, what target segment was chosen, and whether the campaign was launched or left as a draft. That record makes later debugging much easier.

If a tool call fails, do not immediately retry with random parameters. Read the structured error, identify whether the problem is authentication, missing workspace setup, invalid tool arguments, or campaign readiness, and then make the smallest correction. Good MCP usage feels like operating software with a careful assistant, not like repeatedly asking a chatbot to guess.

Empowering Your AI Agents with Context

Connecting your AI agent to a hosted MCP server is the most efficient way to scale B2B outreach while keeping your campaigns relevant. By documenting tools, securing tokens, and configuring pacing limits, you can build a highly automated GTM engine.

Use Omentir to power your integration. Connect your client via MCP, review drafts daily, and launch safe, paced campaigns that turn warm LinkedIn leads into customer conversations.

The teams that get the most from MCP will not be the ones that automate the most actions. They will be the ones that give agents clean context, narrow permissions, and explicit review points. That is how you get the leverage of AI without losing control of the sales motion.

Frequently Asked Questions