Connect AI agents with MCP

ReputeCast includes a Model Context Protocol (MCP) server, so an AI assistant can read reports, check Send Readiness, draft campaigns or pause a send in your workspace: an MCP server email marketing teams can control. Each MCP agent gets its own token and permissions, never more than the person who created it, and anything that would send mail goes to a person as a proposal instead of running.

Before you begin

• You need mcp:manage in the workspace (a sensitive permission). You can only give an agent permissions you hold yourself.
• Your instance needs a public URL your MCP client can reach over HTTPS. The endpoint is <public URL>/mcp, served by the web role.
• If ReputeCast runs behind a reverse proxy, /mcp must stream responses without buffering. See Reverse proxy and TLS.

Create an agent

1. Open Settings › , then Keys & MCP. The MCP card at the top shows the endpoint, for example https://reputecast.example.com/mcp.

2. Select Add agent. Give the agent a name that says who uses it, such as “Claude: analytics”.

3. Pick its permissions. Start with read access, such as reports:view and deliverability:view, and add more later. A permission you don’t hold is locked. Agents can never hold subscribers:verify, placement:run, placement:manage, integrations:manage, ai:use or the agency permissions.

4. Optionally limit the agent to some lists and set an expiry, then create it.

5. Copy the token, rc_agent_…. It is shown once; ReputeCast keeps only its SHA-256 hash.

The agents matrix shows what each agent can do, derived from its permissions:

Column	Allow when the agent holds	Notes
Read reports and contacts	reports:view or subscribers:view	Shows “Allow · provider health” or “Allow · deliverability” for narrower read access.
Draft campaigns	campaigns:edit	Drafts only; nothing is sent.
Edit audience	subscribers:edit, segments:edit or lists:manage
Schedule or send	campaigns:send shows “Approve”; campaigns:pause alone shows “Pause only · never start”	Scheduling always needs a person. See Proposals.
Change routing	providers:manage

Connect your MCP client

The server speaks MCP’s Streamable HTTP transport. Every request carries the agent token as a bearer token:

Request headers

Authorization: Bearer rc_agent_…

ReputeCast accepts only agent tokens here: an API key (rc_live_…) or a session cookie is refused. Pick your client below. The examples read the token from the REPUTECAST_AGENT_TOKEN environment variable; replace reputecast.example.com with your public URL.

Claude Code

Terminal

claude mcp add --transport http reputecast https://reputecast.example.com/mcp \
  --header "Authorization: Bearer $REPUTECAST_AGENT_TOKEN"

To share the server with a project, add it to .mcp.json at the project root instead. Claude Code needs "type": "http" on a server with a url; an entry without type is read as a local command and skipped:

.mcp.json

{
  "mcpServers": {
    "reputecast": {
      "type": "http",
      "url": "https://reputecast.example.com/mcp",
      "headers": { "Authorization": "Bearer ${REPUTECAST_AGENT_TOKEN}" }
    }
  }
}

Run claude mcp get reputecast to check the connection. Source: Claude Code MCP documentation, checked October 2026.

Cursor

Add the server to ~/.cursor/mcp.json (every project) or .cursor/mcp.json (one project):

~/.cursor/mcp.json

{
  "mcpServers": {
    "reputecast": {
      "url": "https://reputecast.example.com/mcp",
      "headers": { "Authorization": "Bearer ${env:REPUTECAST_AGENT_TOKEN}" }
    }
  }
}

Cursor fills ${env:NAME} from your environment. Source: Cursor MCP documentation, checked October 2026.

VS Code

Add the server to .vscode/mcp.json in your workspace. VS Code asks for the token once and stores it securely:

.vscode/mcp.json

{
  "inputs": [
    {
      "type": "promptString",
      "id": "reputecast-agent-token",
      "description": "ReputeCast agent token (rc_agent_…)",
      "password": true
    }
  ],
  "servers": {
    "reputecast": {
      "type": "http",
      "url": "https://reputecast.example.com/mcp",
      "headers": { "Authorization": "Bearer ${input:reputecast-agent-token}" }
    }
  }
}

Source: VS Code MCP configuration reference, checked October 2026.

Other clients

Any client that connects to a remote MCP server by URL and lets you set a request header works. Most use this shape; check your client’s documentation for the file name and whether it also needs "type": "http":

mcp.json

{
  "mcpServers": {
    "reputecast": {
      "url": "https://reputecast.example.com/mcp",
      "headers": { "Authorization": "Bearer rc_agent_…" }
    }
  }
}

A client that only starts local (stdio) servers needs a local bridge to a remote HTTP server; ReputeCast doesn’t ship one.

Treat the token like a password

Anyone with the token can act as the agent. Keep it out of shared repositories: use an environment variable or your client’s secret prompt, as above. If it leaks, revoke the agent in Settings › , then Keys & MCP.

What an agent can do

The server offers 53 tools, 3 resources and 2 prompts. Each tool checks one permission, exactly like an API key, and campaigns can be named by their name or ID. The full list, with each tool’s permission and effect, is in MCP tools, resources and prompts.

• Read: campaign lists and reports, Send Readiness, live send statistics, provider health, routing, deliverability, placement tests, automations, digests, polls, growth, revenue and website content.
• Change: create campaign, SMS, push, WhatsApp and website post drafts; update a campaign’s content; create segments; add addresses to the suppression list; change routing weights; pause and resume sends.
• Send a test: send_test_email sends a campaign test only to members of the workspace, through the System stream.
• Transactional email: read-only. Agents list and render templates and search the message log, but never send transactional messages.
• Resources: a campaign’s MJML, a report as CSV, and the latest 100 audit log entries.

An agent’s permissions are checked again on every call against its creator’s current role. When the creator loses a permission, the agent loses it too; when the creator leaves the workspace or is deactivated, the agent can do nothing.

Proposals for anything that sends

Agents don’t start sends.

• schedule_campaign never schedules. It records a proposal for the first member who holds campaigns:send, who gets a notification (“Claude: analytics asked to schedule Weekly Digest #142”) with a link to the campaign. That person schedules it in ReputeCast, where Send Readiness and approvals apply as usual.
• A call the agent lacks the permission for is denied. The agent gets an error that names the missing permission, the denial is recorded in the audit log, and a proposal goes to the first member who holds it.
• The notification states when the proposal expires: 2 hours by default, 1–168 hours under Change limits. ReputeCast never carries out a proposal itself, so the expiry tells the person how long the request is current; nothing happens when it passes.
• pause_campaign needs campaigns:pause and pauses at once. resume_campaign needs campaigns:send and resumes at once: give campaigns:send to an agent only if it may restart a paused send.

Audit and limits

Every tool call lands in the audit log with the agent as actor, the tool name and the result, including denied calls. Recent agent actions on the same page shows the latest calls; open the audit log for the rest.

Change limits sets the calls per minute per agent and the proposal expiry for the workspace, and turns the MCP server off. While it is off, every agent token is refused. An agent over its limit gets 429 with Retry-After: 60.

Setting	Default	How to change
Calls per minute per agent	120	Settings › Keys & MCP › Change limits, 1–6,000
Proposal expiry stated to the agent and the approver	2 hours	Settings › Keys & MCP › Change limits, 1–168 hours
Idle MCP session ends after	30 minutes	Fixed

Check that it worked

• Ask your assistant to list the tools of the reputecast server, or to call list_campaigns. It should list your campaigns.
• In Settings › , then Keys & MCP, the agent’s last call appears in the agents matrix and under Recent agent actions.

Troubleshooting

• 401 “use an MCP agent token (rc_agent_…)”: the client sends an API key or no header. Create an agent and use its token.
• 401 “this agent was revoked” or “… expired”: create a new agent or rotate its token in Settings › , then Keys & MCP.
• 401 “the MCP server is off”: turn it on under Change limits.
• The client connects but every tool is denied: the agent lacks the permission, or its creator no longer holds it. The error names the permission.
• Claude Code skips the server: the .mcp.json entry has a url but no "type": "http".
• Calls hang or time out behind a proxy: the proxy buffers /mcp responses. Turn off buffering for that path.
• 429 errors: the agent made more calls than the workspace allows per minute. Raise the limit or slow the agent.

Related

MCP tools, resources and promptsEvery tool with its permission and effect.

Roles and permissionsThe permission model agents share with members and keys.

Authentication and API keysKeys for scripts and services that call the REST API.

Use the AI assistant with your own modelThe assistant built into the composer, for people.