Retently MCP Server

Connect Retently to Claude and other AI clients· Back to retently.com

Overview

The Retently MCP server lets an AI client such as Claude work directly with your Retently account in natural language. Ask for your latest NPS, search and tag feedback, reply to detractors, send a transactional survey, or pull a monthly report, without leaving the chat.

It implements the Model Context Protocol over Streamable HTTP, so any MCP-capable client can connect to it as a remote server. There is nothing to install and nothing to host on your side.

Endpoint
https://mcp.retently.com/mcp
Transport
MCP Streamable HTTP (POST to the endpoint)
Authentication
OAuth (recommended) or a personal access token
Tools
53 tools (43 read, 9 write)

What you need

  • A Retently account you can log in to.
  • An AI client that supports remote MCP servers: Claude Code, Claude Desktop, or claude.ai on a Pro, Max, Team, or Enterprise plan.
  • Tools act on your account and respect your Retently role and permissions, so you only ever see and change what your account allows.

Connect with OAuth (recommended)

With OAuth there is no token to generate. You log in once in the browser and approve the access you want. After that the client keeps itself connected in the background, so you never paste a credential. This is the best option for interactive clients.

Claude Code

Add the server, then trigger the login from the REPL:

terminal
claude mcp add --transport http retently https://mcp.retently.com/mcp

In the Claude Code REPL, run /mcp. Your browser opens to Retently, you log in and approve the requested access, and Claude Code stores and renews the session for you. claude mcp list then shows retently as connected.

claude.ai (web)

Go to Settings, then Connectors, then Add custom connector. Name it Retently and use the URL https://mcp.retently.com/mcp. The web app runs the login and approval for you. On the web this is the only connection method, since there is no place to paste a token.

Claude Desktop

Add it as a remote connector under Settings, then Connectors, using the same URL. It runs the same browser login. Restart the app fully after adding it.

How long the session lasts. The client uses a short-lived access token and refreshes it silently in the background. You stay connected as long as you use the connection at least once every 30 days, up to a maximum of 90 days from when you first logged in. After that you log in again.

Connect with a personal access token

A personal access token (PAT) is a long-lived bearer token you generate and paste into a client. Use it for scripts, headless automation, or any client where you want a token you control by hand.

1. Create the token

  • Log in to app.retently.com.
  • Open Settings, then AI Connections.
  • Create a personal access token, choose its scopes (see Scopes and PII), and set an expiry. The default is 90 days and the maximum is 365 days.
  • Copy the token right away. It is shown only once. If you lose it, delete it on the same page and create a new one.

The token looks like this:

rtly_mcp_pat_<id>_<secret>

2. Add it to your client

Claude Code:

terminal
claude mcp add --transport http \
  retently https://mcp.retently.com/mcp \
  --header "Authorization: Bearer rtly_mcp_pat_<id>_<secret>"

Claude Desktop, by editing the config file and then restarting the app fully:

claude_desktop_config.json
{
  "mcpServers": {
    "retently": {
      "type": "http",
      "url": "https://mcp.retently.com/mcp",
      "headers": { "Authorization": "Bearer ${RETENTLY_MCP_PAT}" }
    }
  }
}

Keep the token out of the file by setting it as an environment variable, for example export RETENTLY_MCP_PAT="rtly_mcp_pat_...". On the claude.ai web app there is no place to paste a token, so use OAuth there.

Scopes and PII

Each connection holds a set of scopes that decide what it can do. Grant the least you need. What you can grant is capped by your Retently role.

feedback:readfeedback:tagfeedback:topicfeedback:curatefeedback:deletefeedback:replyfeedback:importcustomers:readcustomers:writecampaigns:readcampaigns:writereports:readsurveys:sendsuppressions:readsuppressions:writetemplates:readoutbox:readoutbox:writeintegrations:readteam:readpii:readpii:write

Reading or changing customer personal data (emails and names) is gated separately by the pii:read and pii:write scopes. These are admin controlled and off by default. Without them, tools still work but return de-identified data. An administrator enables them under Customer PII access on the same AI Connections page.

Tool catalog

53 tools grouped by area. Read tools never change anything. Write tools change data and follow the safety flow described below. Your AI client discovers each tool's arguments automatically once connected.

Feedback

list_feedbackreadList recent feedback, newest first, with score, comment, tags, and topics.
list_recent_feedbackreadList the most recent feedback, optionally within the last N days.
get_feedbackreadGet one feedback record by id, with comment, tags, and topics. Customer identity needs pii:read.
get_feedback_repliesreadGet the reply thread for a feedback record. Sender identity needs pii:read.
search_feedbackreadSearch feedback by comment text, campaign, metric, and score range.
aggregate_feedbackreadAggregate counts and average score grouped by campaign, metric, channel, score, topic, tag, or company.
list_statusesreadList the account's custom feedback workflow statuses.
tag_feedbackwriteAdd or remove tags on a feedback response.
set_topicswriteReplace the full set of topics on a feedback response.
star_feedbackwriteStar or unstar a feedback response for follow-up.
assign_feedbackwriteAssign a feedback response to a team member, or unassign it.
resolve_feedbackwriteResolve or reopen a feedback response, or set its custom status.
add_notewriteAppend an internal note to a feedback response.
reply_feedbackwriteSend a reply email to the customer who left a feedback response.

Filters

describe_filtersreadDescribe the account's feedback filter vocabulary, used to build a filter spec.
apply_filterreadApply a filter spec to account feedback and return rows or an aggregate.
list_saved_filtersreadList the account's named saved filters, each replayable with apply_filter.
list_recent_filtersreadList your recently applied feedback filters, each replayable with apply_filter.

Reports and scores

get_scorereadGet the latest aggregate score (NPS, CSAT, or CES) over an optional date window.
get_monthly_statsreadGet a metric's monthly time series over the last 12 months.
get_topic_statsreadGet per-topic volume, sentiment split, and period-over-period change.
get_reportreadGet delivery stats and score trend for one campaign or all campaigns.
get_trendsreadList the account's trend groups.
get_trends_groupreadGet one trend group's config by id, or the default group.

Customers

list_customersreadList customers, newest first. Email and name require pii:read.
search_customersreadSearch customers by company, tag, or name and email (name and email search needs pii:read).
get_customerreadGet one customer by id. Email and name require pii:read.
update_contactwriteUpdate a contact's profile fields, tags, and custom properties.

Companies

Available on accounts with the B2B Account-CX feature.

list_companiesreadList companies with firmographics and CX metrics.
get_companyreadGet one company by id, with firmographics and CX metrics.

Campaigns

list_campaignsreadList the account's campaigns (name, metric, enabled and default state).
search_campaignsreadSearch campaigns by name, metric, and enabled state.
get_campaignreadGet one campaign by id, with its config and latest aggregate score.

Surveys

send_surveywriteQueue a transactional survey to a list of recipients through an active campaign. Sends live.

Templates

list_templatesreadList survey templates (name, metric, channel, subject).
search_templatesreadSearch survey templates by name, metric, and channel.

Digests

list_digestsreadList generated digest reports, newest first, with scores and report links.
get_digestreadGet one digest report by id: scores, deltas, themes, AI summary, and HTML and PDF links.

Delivery and queue

list_outboxreadList outbound surveys with delivery state. Recipient email needs pii:read.
get_outbox_statsreadGet delivery totals (sent, opened, responded, bounced, opted out).
get_queue_statusreadGet the transactional-survey queue depth.

Suppressions

list_suppressed_domainsreadList suppressed email domains and patterns.
list_suppressed_emailsreadList suppressed email addresses. Needs suppressions:read and pii:read.
get_suppressed_emailreadGet one suppressed email record. Needs suppressions:read and pii:read.

Integrations

list_integrationsreadList integrations with connection state and health. No credentials are returned.
get_integration_statusreadGet one integration's health and last-sync state. No credentials are returned.

Team

list_usersreadList team members with roles and status. Email and name need pii:read.
get_userreadGet one team member by id, with roles and status. Email and name need pii:read.

Account

get_plan_usagereadGet the account plan with usage versus limits per metered dimension.
list_account_attributesreadList the account's custom contact-attribute schema.
list_tagsreadList the customer tags in use on the account.
list_topicsreadList the feedback topics defined for the account.

Server

pingmetaLiveness probe. Returns pong and a server timestamp. No auth required.

Write safety

Every write tool runs as a dry run first. It returns a preview of exactly what would change plus a confirmation token, and changes nothing until the client calls it again with that token. Write actions are rate limited per account and recorded in an audit log.

  • send_survey sends real surveys to real customers once confirmed. Use a test contact while you are trying it out. Suppressed addresses are dropped automatically.
  • reply_feedback sends a real email to the customer. Sending cannot be undone.

Verify it works

  • In Claude Code, /mcp should show retently connected with a tool count above zero.
  • Ask the client to call ping. It returns { "pong": true } and proves the connection is reachable, even without a valid credential.
  • Ask for your most recent feedback or your NPS score. You should get rows scoped to your account.

The same check from the command line with a token:

terminal
curl -s https://mcp.retently.com/mcp \
  -H "Authorization: Bearer rtly_mcp_pat_<id>_<secret>" \
  -H "Content-Type: application/json" \
  -H "Accept: application/json,text/event-stream" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"get_score","arguments":{}}}'

Troubleshooting

SymptomWhat to do
The browser login never opensIn Claude Code, run /mcp to start the flow, and make sure you added the server without an Authorization header. On the web, use the Connectors screen, which drives the login for you.
Every data tool says it is not authenticatedThe login did not complete, or the token is missing, expired, or sent incorrectly. Log in again, or create a new token and check the header is Authorization: Bearer <token>.
ping works but reads do notThe connection is fine and the problem is the credential. Re-authenticate with OAuth, or create a new token.
A tool returns a permission errorYour connection is missing that scope, or your role or PII settings do not allow it. Add the scope when you create the token, or ask an administrator.
No tools appear after connectingRestart the client. Claude Desktop needs a full restart.

Looking for the REST API instead? See the Retently API reference.