MCP server
ForgeSend runs a Model Context Protocol (MCP) server, which lets an AI agent — like Claude or another MCP-compatible assistant — act on your ForgeSend workspace directly, instead of you clicking through the app yourself.
On this page
What is MCP
MCP is an open standard that lets an AI assistant call a defined set of tools against a real system — in this case, your ForgeSend workspace — rather than just talking about it. Once connected, an agent can list your campaigns, add contacts to a list, pull analytics, and more, using the same underlying actions the ForgeSend app itself uses.
Every action an MCP tool takes is scoped to the single workspace the connecting API key belongs to. An agent connected via MCP can't see or act on any other workspace.
Authentication
There are two ways to connect an MCP client to your workspace.
Workspace API key
MCP clients authenticate with an API key, generated from Settings → API.
API Keys
Generate keys for external tools — like the MCP server — to access this workspace without a browser login.
Generating a key requires being a workspace owner. The full key is shown exactly once, immediately after generation — copy it right away, since it can't be displayed or recovered afterward, only revoked and replaced with a new one.
Give the key a descriptive label like "MCP server" so it's easy to identify — and revoke — later from the Active keys list without guessing which key is used where.
Configure your MCP client with that key as a bearer token against ForgeSend's MCP endpoint. Requests are rate-limited per workspace, so a single client behaving normally won't run into limits in everyday use. A key's scopes decide which of the 24 tools below it can actually call — a key without sending:write, for instance, can't be used to start or pause a campaign.
OAuth 2.1 via a custom connector
The other option is connecting through ForgeSend's own OAuth 2.1 authorization server — no key to generate, copy, or rotate by hand. This isn't a listing you find in a public connector directory: in an MCP client like Claude, you add ForgeSend as a custom connector yourself, pointing it at ForgeSend's MCP endpoint. From there, the client registers itself with ForgeSend automatically (dynamic client registration) and starts the OAuth handshake — nothing needs to be pre-approved or submitted anywhere on ForgeSend's side first.
That handshake lands you on a consent screen, after signing in if you weren't already.
ForgeSend
Authorize Claude
Claude wants to connect to your ForgeSend workspace using the Model Context Protocol.
Workspace
This will allow Claude to:
Signed in as alex@meridian.co
This isn't a single-click connection — it's login (if needed), then this explicit consent screen, with an Approve/Deny choice you make every time a new client requests access. If you belong to more than one workspace, a workspace picker appears above the permissions list, since consent is granted for exactly one workspace at a time. The permissions list is all-or-nothing — there's no way to approve some requested scopes and deny others on this screen.
Whichever method you use, both are fully supported side by side — OAuth didn't replace the workspace API key, it's simply the option that doesn't require you to manage a key yourself. The scopes an OAuth connection can be granted are the exact same six resources, twelve values an API key can be scoped to.
The three tool categories
The MCP server exposes 24 tools across three areas:
Campaign management
Create, update, schedule, start, pause/resume, review held jobs, and manage sequence steps
Lists / Contacts
View lists, add/update/search contacts, enroll contacts into a campaign, read reply content
Analytics
Read-only — workspace, per-campaign, per-step, and per-variant stats, plus daily send volume
- Campaign management — an agent can list your campaigns, create new ones, rename them, update their schedule, start a campaign that's still in Draft, pause or resume sending, add, edit, or remove sequence steps, and approve or reject pre-send jobs that have been held for review.
- Lists / Contacts — an agent can list your contact lists, add or update contacts, search contacts by name, email, or company, add an entire list to a campaign as leads, and read the actual content of a reply conversation — not just its intent tag. See Reply Box for how replies and intent tagging work.
- Analytics — strictly read-only. An agent can pull workspace-wide analytics, per-campaign stats, per-step and per-A/B-variant performance breakdowns, and recent daily send volume, but has no tool that creates, changes, or deletes anything in Analytics.
What's intentionally not exposed
Enrichment and Find Leads have no MCP tools at all — an agent connected via MCP cannot run a lead search, enrich a contact, or spend enrichment credits in any way.
This is a deliberate boundary, not a gap to be filled later: Find Leads and Enrichment both spend real credits per action. Keeping them out of MCP protects against an agent unintentionally burning through your enrichment or lead-search balance on your behalf.
If you want an agent to work with newly found leads, the workflow is to run Find Leads yourself, then hand the resulting list to the agent — from there, its Lists/Contacts and Campaign management tools can take over.
Where to go next
Next: Getting started