myagentmail-mcp

Model Context Protocol server for myagentmail. Lets Claude Desktop, Cursor, Windsurf, Cline, Continue, and any other MCP-compatible client drive both email and LinkedIn workflows against the myagentmail API:

• Email — provision inboxes, send / reply / forward, work with threads, drafts and attachments, register custom domains, manage workspaces and webhooks.
• LinkedIn — connect accounts, send connection requests + DMs, read inbox + threads, list 1st-degree connections + sent invitations, publish posts (text + image), search content, and create recurring intent signals (keyword, engagement, watchlist, connection_accepted, message_received) with webhook delivery.
• Cadences — server-side outreach sequences. Define an ordered list of steps (LinkedIn invite → wait → DM → email → ...), enroll leads, and the platform runs them end-to-end with branching on real-world events (invite accepted, lead replied). The runtime is on MyAgentMail's side; your app just calls the API and listens for webhook events.

Install

npm install -g myagentmail-mcp

Or use with npx (no install required):

npx myagentmail-mcp

Install the Claude skill (optional but recommended)

The MCP server gives a Claude / Cursor / Windsurf agent the tools. The Claude skill gives it the playbook — auth model, capability map, working recipes, and the deliberate omissions (we don't ship enrichment or ICP search). With both installed, an agent self-orients and starts shipping correct code on its first try.

One command drops SKILL.md and the three reference docs into a Claude skills directory:

# In the project you're working in:
npx myagentmail-mcp install-skill

# Or globally for every Claude session on this machine:
npx myagentmail-mcp install-skill --global

This writes to ./.claude/skills/myagentmail/ (project) or ~/.claude/skills/myagentmail/ (global). Add --force to overwrite an existing install.

Setup

You need a myagentmail API key. Get one from https://myagentmail.com/dashboard.

The server reads it from the MYAGENTMAIL_KEY environment variable. Each MCP client has its own way of injecting env vars — see below.

Claude Desktop

Edit ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "myagentmail": {
      "command": "npx",
      "args": ["-y", "myagentmail-mcp"],
      "env": {
        "MYAGENTMAIL_KEY": "tk_your_key_here"
      }
    }
  }
}

Restart Claude Desktop. The myagentmail tools appear in the tool picker.

Cursor

Edit ~/.cursor/mcp.json (or your project's .cursor/mcp.json):

{
  "mcpServers": {
    "myagentmail": {
      "command": "npx",
      "args": ["-y", "myagentmail-mcp"],
      "env": { "MYAGENTMAIL_KEY": "tk_your_key_here" }
    }
  }
}

Windsurf

Edit ~/.codeium/windsurf/mcp_config.json:

{
  "mcpServers": {
    "myagentmail": {
      "command": "npx",
      "args": ["-y", "myagentmail-mcp"],
      "env": { "MYAGENTMAIL_KEY": "tk_your_key_here" }
    }
  }
}

Cline / Continue / other MCP clients

Same pattern: configure a server named myagentmail, command npx, args ["-y", "myagentmail-mcp"], env {"MYAGENTMAIL_KEY": "..."}.

Tools exposed

This server wraps the most useful subset of the myagentmail REST API. The LLM sees a JSON Schema for each tool and can call it directly via MCP.

Inboxes

• create_inbox — provision a new mailbox with IMAP/SMTP credentials
• list_inboxes — list all inboxes
• get_inbox — fetch one inbox with its address list and credentials
• delete_inbox — soft-delete

Messages

• send_message — send a new email
• reply_to_message — reply in-thread (preserves headers)
• reply_all_to_message — reply-all, dropping self addresses
• forward_message — forward to new recipients
• list_messages — paginated listing, filterable by direction
• get_message — full body + attachment metadata
• mark_message_read — flip the read flag
• delete_message — soft-delete

Threads

• list_threads — paginated listing
• get_thread — every message in a thread, ordered

Drafts

• create_draft — start a draft with partial fields
• update_draft — merge-patch any field
• send_draft — send and atomically delete the draft
• list_drafts — list unsent drafts
• delete_draft — discard a draft

Attachments

• list_attachments — metadata for a message
• download_attachment — returns base64-encoded bytes

Custom domains

• register_domain — register and get DNS records to set
• verify_domain — re-check DNS status
• list_domains — list registered domains
• delete_domain — remove
• add_inbox_address — attach a verified custom-domain alias to an existing inbox

Workspaces (multi-tenant)

• list_workspaces — list isolation containers in your tenant
• create_workspace — provision a new workspace
• issue_workspace_key — mint a wk_ key scoped to one workspace

Webhooks

• create_webhook / list_webhooks / delete_webhook

Metrics

• get_metrics — sent / received / bounced counts and resource totals

Utility

• verify_email — pre-flight MX check on one or more recipient addresses

LinkedIn — Sessions

• linkedin_login — start email/password login (may return a challenge)
• linkedin_verify_pin — submit the email/SMS PIN if challenged
• linkedin_poll_mobile_approval — collect a mobile-app approval challenge
• linkedin_import_session — import li_at + JSESSIONID cookies (escape hatch)
• linkedin_list_sessions / linkedin_get_session / linkedin_revoke_session
• linkedin_sessions_utilization — per-session daily quota usage

LinkedIn — Outreach

• linkedin_lookup_profile — resolve a profile URL to its profileId, name, headline
• linkedin_send_connection_request — send an invite (with optional note)
• linkedin_list_connections — list 1st-degree connections
• linkedin_list_sent_invitations — list pending outbound invites
• linkedin_send_message — DM a 1st-degree connection (find-or-create thread)
• linkedin_list_conversations — list inbox threads
• linkedin_read_conversation — read messages in a thread

LinkedIn — Posts

• linkedin_create_post — publish a feed post (text + optional image)
• linkedin_list_posts_by_author — recent posts by a profile / company
• linkedin_list_post_comments — commenters + bodies
• linkedin_list_post_reactions — reactors + reactionType
• linkedin_search_content — raw keyword post search

LinkedIn — Historical search (with classifier)

• linkedin_search_history — one-shot historical search with firing rule + classifier verdict per result
• linkedin_search_history_list / linkedin_search_history_get

LinkedIn — Signals (real-time intent watchers)

• linkedin_signal_create — keyword | engagement | job_change_watchlist | connection_accepted | message_received
• linkedin_signal_create_engagement / linkedin_signal_create_watchlist — kind-specific helpers
• linkedin_signal_list / linkedin_signal_get / linkedin_signal_update / linkedin_signal_delete
• linkedin_signal_run — force-poll now
• linkedin_signal_matches — historical match list
• linkedin_signal_deliveries — webhook delivery log
• linkedin_signal_test_webhook — fire a synthetic test event
• linkedin_signal_watchlist_list / linkedin_signal_watchlist_add / linkedin_signal_watchlist_remove — manage watchlist signal entries

Cadences — server-side outreach sequences

• cadence_create — create with optional inline steps array
• cadence_list / cadence_get / cadence_update / cadence_delete
• cadence_replace_steps — atomically replace the step list
• cadence_enroll_lead — enroll a lead (idempotent on (cadenceId, leadExternalId))
• cadence_list_enrollments — filter by state (active / paused / completed / replied / ...)
• cadence_run_now — force a runner tick (debug)
• enrollment_get — one enrollment + step-run history
• enrollment_patch — pause / resume / unenroll

Key tiers

myagentmail has four key types — pick the narrowest that fits the job:

| Prefix | Scope | Use | |---|---|---| | tk_ | Tenant master — every workspace | Server-side admin, dev | | wk_ | Workspace master — one workspace | Reseller customers | | ak_ | Inbox scoped — one inbox | Production agent runtimes | | mam_ | Legacy inbox key | Backwards compat |

For local Claude Desktop / Cursor use, a tk_ key is fine. For deployed agents, prefer ak_ to limit blast radius.

Develop locally

git clone https://github.com/kamskans/myagentmail
cd myagentmail/mcp
npm install
npm run build

# Test interactively
MYAGENTMAIL_KEY=tk_... node dist/index.js

The server speaks JSON-RPC over stdio. Use the MCP Inspector for interactive testing during development.

Architecture

This package is a thin wrapper. The actual API lives at https://myagentmail.com/v1 — every tool here just translates an MCP tool call into a fetch against the matching REST endpoint and returns the JSON response.

If you don't want the MCP layer, you can call the same endpoints directly:

curl -H "X-API-Key: $MYAGENTMAIL_KEY" https://myagentmail.com/v1/inboxes

See the full API reference and the knowledge base for concept guides.

SDK

For programmatic use in your own code, install the SDK instead: npm install myagentmail (npm).

License

MIT