MCP ServerNEW

InboxPlease MCP Server Deliverability Tools for AI Agents

Connect Claude, ChatGPT and any MCP client to InboxPlease. Agents can monitor warmup and pause or resume it, run inbox placement tests, analyse emails, read Domain Health, get a deliverability overview, analyse domains and send email (opt-in).

Available on: Enterprise
terminal
# Connect InboxPlease to Claude Code
claude mcp add --transport http inboxplease \
  https://api.inboxplease.com/mcp \
  --header "Authorization: Bearer inp_YOUR_API_KEY"
Summary

What is MCP Server?

The InboxPlease MCP server is a Model Context Protocol endpoint that lets AI agents call InboxPlease tools using a scoped API key: a deliverability overview, Domain Health results and scan history, warmup status and stats with pause and resume, inbox placement tests (manual or from a connected mailbox), email analysis, domain analysis and profiles, reputation and SEO/GEO metrics, mailbox tools, and opt-in sending and message tools. Every developer scope is opt-in. Tools that change something tell the agent to confirm with you first, sending is a separate owner/admin scope, and every send is staged as a preview that needs explicit confirmation.

Protocol
MCP over Streamable HTTP
Clients
Claude, ChatGPT, any MCP client
Tools
Overview, warmup, placement, analysis, domains, mailboxes, email
Auth
inp_ API key with /mcp/ scope
Limits
300 requests/min · 10 concurrent calls per key
Sending
Opt-in scope · two-phase preview → confirm
Plan
Enterprise

Safe by default

Agents get exactly the access you grant — nothing more.

Scoped keys

Restrict each key to specific tools and sender addresses.

Send is opt-in

The send scope must be explicitly requested at key creation. It's never granted by default.

Preview before send

send_email stages a preview and single-use token. Nothing is sent until confirm_send.

Recipient quotas

Per key, per mailbox and per organization limits — max 5 recipients per send.

Human in the loop

Connecting a mailbox returns an OAuth link a person must open.

Audited

Every staged send is recorded with each state change.

Owner/admin for actions

Sending, connecting mailboxes, pausing warmup and starting placement tests need a key created by an owner or admin.

Prompt-injection safeguard

Content an attacker could control (email bodies, headers, subjects) is labelled as untrusted in tool output.

Per-key rate limits

300 requests/min and at most 10 calls running at once per key. Every call is recorded per key.

Built with InboxPlease: Outreach Please

Outreach Please is a reference outreach app built only on the public REST API and this MCP server. Its server holds the inp_ key, stages every send and confirms it only after a person approves the preview, and checks Domain Health before a campaign goes out, with an optional placement test.

See how Outreach Please is built
InterfacesAuth REST API + MCP
KeysServer-side only
SendingPreview → human approval → confirm

Tools agents can call

whoamiWhich organization, plan and scopes this key has
get_deliverability_overviewDeliverability overview with recommendations
list_health_domains · get_domain_healthMonitored domains and their latest health
get_domain_scan_history · get_domain_scanScan history and one scan's result
list_warmups · get_warmup_statsWarmup status, inbox rate and per-sender stats
pause_warmup · resume_warmupPause or resume warmup (asks you first)
start_placement_testInbox placement test, manual or from a connected mailbox (asks you first)
get_placement_result · list_placement_testsInbox / tabs / spam per provider
analyze_emailAuthentication, sending IP, hops and content verdicts
domain_analysis · domain_profileDomain deliverability analysis and full profile
reputation_scores · seo_geo_analysisReputation signals and SEO/GEO metrics
list_mailboxes · check_mailbox_connectionConnected mailboxes and connection status
connect_mailboxOAuth link a person must open
get_send_quotaRemaining recipient budget
send_email → confirm_sendTwo-phase send: preview first, then confirm. Opt-in scope.
get_send_statusCheck a send's outcome by its send_id, without sending again
get_campaign_capabilities · get_send_engagementCampaign readiness, and opens, clicks, replies and bounces for a send
read_email · mark_email_read · star_email · move_emailOne known message by ID, on IMAP-connected mailboxes only. Separate opt-in scope.

Questions about MCP Server

Can AI agents like Claude use InboxPlease?

Yes. On Enterprise, the InboxPlease MCP server lets Claude, ChatGPT and other MCP clients get a deliverability overview, read Domain Health results and scan history, monitor warmup and pause or resume it, run inbox placement tests, analyse emails, run domain analysis, domain profiles, reputation scores and SEO/GEO checks, and check connected mailboxes, all with a scoped API key. Actions need opt-in scopes on an owner/admin key, and the agent confirms with you first. Sending is opt-in, previewed first and audited.

Can an agent pause my warmup or start placement tests?

Only if you allow it. Pausing or resuming warmup and starting placement tests each need an opt-in scope ticked when the key is created, and those scopes only work on a key created by an owner or admin of your organization. The agent is also told to confirm with you before it acts.

Can an agent read my inbox?

No agent can browse or search your inbox, and mailboxes you connect through our standard app are send-only. If you tick the separate read scope on an owner/admin key, read_email can fetch one message whose ID the agent already has, but only from a mailbox connected over IMAP. Its content is labelled as untrusted.

Connect your agent

MCP access is available on Enterprise.