TextKit

npx add-mcp 'https://mcp.jointextkit.com'

MCP server

Source: https://jointextkit.com/docs/mcp Index: https://jointextkit.com/llms.txt

Connect TextKit to Claude, Cursor or any MCP client and the agent can send texts, buy numbers, load contacts, build segments, run broadcasts and set up automations on your account.

The server is https://mcp.jointextkit.com. Sign in to connect, or pass an API key.

Connect from Claude

claude.ai and Claude Desktop: Settings → Connectors → Add custom connector, URL https://mcp.jointextkit.com. Claude sends you to TextKit to sign in. Approve the connection, pick test or live data, and you are back in Claude.

Claude Code:

claude mcp add --transport http textkit https://mcp.jointextkit.com

Then run /mcp in Claude Code and choose TextKit to sign in.

Approving creates an API key named for the app, listed on the API keys page; revoke it there to disconnect. Connecting the same app again replaces its key. Only an account admin can approve.

Connect from ChatGPT

In ChatGPT, open Plugins, select +, then Add custom MCP server, with the URL https://mcp.jointextkit.com/chatgpt and OAuth, and create it as a plugin. ChatGPT sends you to TextKit to sign in; approve it the same way. This URL has every tool except Verify and API keys: ChatGPT does not allow one-time codes.

Step-by-step guides for Lovable, Replit, Bolt, v0, VS Code, Gemini CLI, Cline and more are on the integrations page.

Add to Replit or Cursor

Add to Replit · Add to Cursor

Both open the app with TextKit filled in; sign in when it asks. In Replit, Agent can then text from the apps it builds: ask it to create an API key with create_api_key and store it as the secret TEXTKIT_API_KEY.

Connect with an API key

Any MCP client that can send a header:

{
  "mcpServers": {
    "textkit": {
      "url": "https://mcp.jointextkit.com",
      "headers": { "Authorization": "Bearer ssms_sk_test_..." }
    }
  }
}

Start with test data: everything runs in the sandbox and no real phone gets a text. Every tool is one API call with your key, so its scopes, rate limits and send checks apply unchanged. To limit what an agent can do, give it a key with only the scopes it needs.

Things to ask

Buy a number in area code 512 and text +15005550006 "Your order shipped."

Import the contacts in this CSV, tag them spring-sale, and tell me how many have an opt-in.

Send everyone tagged vip "Doors open at 9 on Saturday." Do a dry run first and show me the count and cost.

When someone triggers cart_abandoned, wait an hour, and text them a reminder unless checkout_completed arrives first.

How did yesterday's broadcast do? Show delivered, failed, replies and opt-outs.

Bring an audience from any tool

Your people can be anywhere: an analytics tool, a CRM, a store, an email tool, billing, your own database, a spreadsheet or a file. Ask for them and the agent works out how to get them.

Import my users who can get marketing texts.

If you have not said where they are, the agent asks (plan_import): where your people live, which group, what you can sign in to, whether consent was recorded, and once or kept in sync. Then it offers the ways in, best first:

Way in	When
Tools already in the chat	The agent has that tool's MCP server, a connector, or database access: nothing to set up
Integration	Shopify, HubSpot, Salesforce, Mixpanel, Amplitude, synced every 6 hours. You type the credentials into the console link it gives, never into the chat
Push	Segment, RudderStack, Hightouch, Census or a cohort webhook posts people as they change
Script	A key you keep in your own environment; the agent writes the code
Export or CSV	The agent tells you where the export button is, then reads the file or you upload it in the console

Bring the "Trial ending this week" cohort from Mixpanel into TextKit and text them a reminder about the upgrade offer.

With a source named, the agent then:

1. Reads the cohort with the other tool: phone, name, its id there, any SMS consent it recorded, and a few attributes worth merging.
2. Calls import_contacts with dry_run: true and shows you what will be created, updated, skipped or already opted out. Up to 500 per call.
3. Imports them tagged (for example mixpanel-trial-ending) and saves a segment on that tag.
4. Calls send_batch with dry_run: true: recipients, cost, and who is removed and why. It sends after you approve.

Run it again later to refresh: contacts are matched by phone number, and the agent removes the tag from people who left the cohort.

Consent comes from the source, never from the agent. A cohort is not permission to text. Marketing goes only to people with an opt-in on file, so the agent carries over opt_in only where the source recorded one (when, and where it was collected). Without it they can still get transactional texts. See opt-out and consent.

Prompts

Three built-in playbooks show up as commands in clients that support MCP prompts (in Claude Code: /mcp__textkit__import_audience):

Prompt	Arguments	What it does
import_audience	source, audience, tag (all optional)	Asks where your people are if you have not said, offers every way in, imports with a dry run first, keeps the tag in sync, saves a segment
send_campaign	audience, goal	Finds or imports the audience, drafts the text, dry-runs it, sends or schedules after you approve
campaign_report	period	Delivery, replies, opt-outs and spend per batch, the trend, and a follow-up cohort

Tools

Area	Tools
Messages	send_message, get_message, list_messages, delete_message, redact_message, simulate_inbound, list_events
Verification	send_verification, check_verification, fallback_verification, get_verification
Numbers	list_numbers, get_number, search_available_numbers, purchase_number, bulk_purchase_numbers, bulk_update_numbers, release_number, link_number_to_registration, assign_number_to_customer, export_numbers, lookup_phone
Sender pools	list_number_pools, create_number_pool, get_number_pool, update_number_pool, delete_number_pool, clone_number_pool
Contacts	plan_import, import_contacts, upsert_contact, list_contacts, get_contact, update_contact, delete_contact, bulk_update_contacts, set_contact_topic
Segments and topics	query_audience, create_segment, list_segments, get_segment, update_segment, delete_segment, preview_segment, list_topics, create_topic, get_topic, update_topic, delete_topic, get_contact_topics
Consent	get_consent, set_consent, list_opt_outs, import_opt_outs
Templates	create_template, list_templates, get_template, update_template, delete_template
Broadcasts	send_batch, list_batches, get_batch, get_batch_results, list_batch_recipients, pause_batch, resume_batch, cancel_batch, retry_batch
Scheduled and recurring	list_scheduled_messages, get_scheduled_message, cancel_scheduled_message, create_schedule, list_schedules, get_schedule, update_schedule, delete_schedule
Automations	track_event, create_automation, list_automations, get_automation, update_automation, delete_automation, activate_automation, pause_automation, list_automation_runs, get_automation_run, cancel_automation_run
Account	get_balance, get_usage, get_spend_limit, set_spend_limit, get_deliverability, list_audit_logs, list_registrations, get_registration, list_registration_submissions, list_toll_free_verifications, get_toll_free_verification, forget_identity
Reports and exports	export_csv (messages, deliveries, events, opt-outs, usage, audit log, credit history), export_consent, wait_for_events
Webhooks	create_webhook, list_webhooks, get_webhook, update_webhook, delete_webhook, rotate_webhook_secret, test_webhook, replay_failed_webhooks, list_webhook_deliveries, replay_event
Customers (platforms)	create_customer, list_customers, get_customer, update_customer, delete_customer, get_customer_usage

Tools that send, delete or turn something on are marked destructiveHint, read-only ones readOnlyHint, so your client can ask before running them. Each tool is one call to the API and takes the same inputs, so a new field or list filter there shows up in the tool too. A failed call returns the API's error object with isError: true. Filing a 10DLC registration or toll-free verification (they carry a business attestation and fees) and fetching the webhook signing secret stay in the console and the API.

Cohorts

Ask about your audience in plain words and the agent turns it into segment rules: "who replied in the last 7 days", "who got Friday's broadcast and didn't answer", "who started checkout but never bought". query_audience answers with the counts and the first contacts without saving anything; create_segment saves the cohort so send_batch can text it. See segments for every rule.

The agent asks before it acts

For anything that needs your details (buying a number, a 10DLC registration, a toll-free verification, a recurring text, a webhook, an automation), plan_task gives the agent the questions to ask you, each with your account's options (your numbers, registrations, topics, segments) and every field the API takes. It asks what you have not already said, then does it. Carrier registrations are drafted with you and submitted by you in the console.

Protocol

Streamable HTTP, stateless JSON (no sessions or SSE), protocol version 2025-03-26. initialize, tools/list and prompts/list work without a key; tools/call needs one. Discovery: /.well-known/mcp.json.

New tools and fields reach a client when it next lists tools. A stateless server cannot push notifications/tools/list_changed, so a client keeps the list it read when it connected: reconnect it (in Claude Code, /mcp and Reconnect) to pick them up. serverInfo.version changes whenever the tool list does, so a client can tell its list is stale.