clevertap-mcp
# clevertap-mcp
An MCP server that gives an LLM agent direct control of CleverTap.
Built so a Claude agent could plan, launch and measure CRM campaigns without a
human opening the CleverTap dashboard. It is the execution rail underneath an
agentic CRM loop running against a ~1.8M user base.
## What it exposes
| Area | Tools |
|---|---|
| Campaigns | `create_campaign`, `stop_campaign`, `get_campaign_report` |
| Send | `send_push`, `send_email`, `send_sms`, `send_whatsapp` |
| Profiles | `get_profile`, `get_profiles`, `get_profile_count`, `upload_profiles` |
| Events | `get_events`, `get_event_count`, `upload_events`, `get_realtime_counts` |
| Analysis | `get_trends`, `get_top_property_values` |
The point of the profile and event tools is that the agent can size a segment
*before* it commits to a send, so cost estimation happens inside the planning
loop rather than after the fact.
## Setup
```bash
npm install
npm run build
```
Environment:
```bash
CLEVERTAP_ACCOUNT_ID=...
CLEVERTAP_PASSCODE=...
CLEVERTAP_REGION=in1 # optional, defaults to in1
```
Register with your MCP client pointing at `dist/index.js`.
No credentials are stored in this repo. Everything reads from the environment.
TDQS
Scored across 19 tools
Each tool targets a distinct resource and action: profiles, events, campaigns, messaging channels, and analytics. Even similar tools like get_profile_count vs get_event_count are clearly differentiated by what they count (users vs events) and time context (live vs historical). No two tools appear to overlap in a way that would cause misselection.
All tool names follow a clear verb_noun pattern with consistent verbs: get, create, stop, send, upload. The object nouns are specific (profile, campaign, event, trends, etc.). Minor variations like get_realtime_counts vs get_event_count do not break the pattern; everything is snake_case and starts with a verb.
19 tools is on the higher end but justified for a customer engagement platform that spans profiles, events, campaigns, messaging channels, and analytics. Each tool addresses a distinct need, though the count could be slightly trimmed (e.g., merging pagination helpers). Still, it feels appropriately scoped and not bloated.
The surface covers the core lifecycle: profile creation/update/retrieval, event upload/query, campaign create/stop/report, and direct messaging across multiple channels. Minor gaps exist, such as no update_campaign or delete_profile/delete_event, but these are not critical for most workflows. Overall, the domain is well covered without dead ends.