Remote MCP Servers in Campaigns

Why It Matters

Kit’s outreach agent already researches each prospect using web search, source URLs, and the campaign knowledge base. Sometimes you need it to also pull in data only your team has — a deal stage from your CRM, a customer health score from your warehouse, an internal playbook from your wiki. Remote MCP servers bridge that gap. Connect any server that implements the Model Context Protocol and the AI agent can call its tools alongside Kit’s built-in ones.

What Is a Remote MCP Server

Model Context Protocol is an open standard for exposing tools to AI agents over JSON-RPC. If you can host an HTTPS endpoint that responds to tools/list and tools/call, the outreach agent can use it. You bring the server, Kit handles the orchestration.

Common use cases:

Use Case	Example
CRM lookup	“Has this prospect interacted with us before?”
Internal documentation	“Find the relevant case study for a fintech buyer.”
Custom data warehouse	“Look up the MRR of this prospect’s company.”
Proprietary scoring	“Score this prospect against our ICP model.”

Requirements

For Kit to connect to a remote MCP server it must be:

• Publicly reachable over HTTPS. HTTP endpoints and private IP ranges are rejected for security. Self-signed certificates are not supported.
• Compatible with JSON-RPC 2.0 as defined by the MCP specification.
• Authenticated by bearer token (optional). If your server requires auth, provide a token when connecting and Kit will send it as Authorization: Bearer <token> on every request.
• Reasonable in size. Kit caps responses at 512 KB and the tool list at 50 tools per server. Each account can connect up to 10 servers.

Connecting a Server

1. Open Integrations > MCP Servers and click New server.
2. Provide a recognizable name (this is what the AI sees in its tool list), the HTTPS endpoint URL, and an optional bearer token.
3. Save. Kit immediately enqueues a tool discovery job that calls tools/list on the server, caches the schema, and marks the server active.

Discovery usually takes a few seconds. Until it completes the server is in the Discovering state and is not yet selectable inside a campaign. If discovery fails, the server moves to Failed with the error message from the server. After three consecutive failures Kit disables the server automatically and notifies the account admin.

Selecting Servers in a Campaign

Open any campaign, edit its configuration, and find the Remote MCP servers section. Every active server in your account appears with a checkbox, its endpoint URL, and the number of tools it exposes. Tick the servers you want this campaign to use.

A few notes:

• Selection is per campaign. Different campaigns can use different server combinations.
• Only active servers appear. Servers still discovering or in a failed state are summarized in a notice below the list.
• Deselecting all is fine. The AI agent will fall back to its built-in tools.
• Tool names are namespaced. When the AI sees a remote tool it appears as mcp__<server_id>__<tool_name>, so two servers with overlapping tool names do not collide.

How the Agent Uses the Tools

During research the AI agent receives the union of Kit’s built-in tools and the tools from every selected server. It picks which tool to call based on the tool’s description, so the quality of your remote tool descriptions directly affects how often the agent uses them. Be specific about what each tool does and what arguments it expects.

Tool calls are logged on the message and surfaced in the draft review UI. You can see exactly what the agent called, with what arguments, and what came back.

Security Model

Remote MCP servers run on your infrastructure under your authentication. Kit acts purely as a client:

• Kit never proxies traffic from prospects, end users, or third parties to your server. Only the outreach agent does, and only while drafting an email for a prospect in a campaign you have configured.
• The bearer token is encrypted at rest and never exposed to the AI model.
• Kit validates that the endpoint hostname resolves to a public IP before each request, defending against SSRF.
• Tool arguments are passed verbatim from the AI, so server-side input validation is your responsibility.

If you have compliance requirements about which tools can be called when, enforce them in your MCP server, not in Kit.

Troubleshooting

Symptom	Likely Cause	What to Try
Server stuck in Discovering	Tool discovery job has not yet run	Wait a minute, refresh the integrations page
Server in Failed with HTTP 401	Bearer token missing or invalid	Edit the connection and update the token
Server in Failed with “Hostname resolves to private IP”	Endpoint points at a private network	Expose your server through a public hostname
Active server reports 0 tools	Server returned an empty tools/list	Confirm your server registers tools and re-trigger discovery
Server is active but the AI never calls its tools	Tool descriptions are too vague	Rewrite descriptions to clearly state what each tool does
Server disabled automatically	Three consecutive failures	Fix the underlying issue, then re-enable the server

Quick Checklist

• Your MCP server is reachable over public HTTPS
• You have created the connection in Integrations > MCP Servers
• The server is in the Active state with at least one tool
• You have selected the server inside the campaign you want it used in
• Tool descriptions are concrete enough for the AI to know when to call them

Next Steps

• MCP Tools Reference — see the built-in tools the AI agent has by default
• AI Research and Knowledge Base — how research and the knowledge base work alongside MCP tools