MCP server: Pushizo with AI assistants
With its Model Context Protocol (MCP) server, Pushizo lets Claude Desktop, Claude.ai, Cursor and other MCP clients read your segment, campaign, message, template, journey, user, analytics and web growth data and…
With its Model Context Protocol (MCP) server, Pushizo lets Claude Desktop, Claude.ai, Cursor and
other MCP clients read your segment, campaign, message, template, journey, user, analytics and web growth data,
and, with your approval, prepare draft campaigns, segments, templates, journeys, in-app messages, web campaigns, web
personalization experiences, surveys and landing pages
(docs/adr/0031-entegrasyonlar-ve-mcp-sunucusu.md §6, docs/adr/0043-mcp-arac-genisletmesi.md).
The assistant can also run the setup: it creates the app, returns the setup code, uploads the push key from a file
without its contents entering the conversation, and verifies the first subscriber
(AI-assisted setup, docs/adr/0097-yapay-zeka-ile-kurulum.md).
The server has no tools for sending to an audience, scheduling, publishing or deleting. The only sending tool is send_test_notification, and
it only reaches the app's test devices. Drafts are reviewed in the dashboard and then sent or published.
There are two ways to use it:
| Mode | When | Credentials |
|---|---|---|
Local stdio (npx -y @pushizo/mcp-server) | Local server in Claude Desktop | PUSHIZO_API_KEY environment variable |
Remote Streamable HTTP (POST {API_URL}/mcp) | Claude.ai / Claude Desktop custom connector, clients that need no installation | Scoped API key (Bearer) or OAuth 2.1 |
The server is listed in the official MCP Registry as com.pushizo/mcp (domain
verification: https://pushizo.com/.well-known/mcp-registry-auth); source:
github.com/pushizo/pushizo-mcp.
In both modes the tools only call the public REST API: scope checks, the API key's IP allowlist, tenant isolation and RLS are exactly the same as the public API on every call.
Tools and scopes
The server has 231 tools: 220 read tools, 9 draft tools, app creation (create_app) and the test send. The local
stdio server also has upload_channel_credentials (232). All of them call the public REST API;
the scopes below are also granted by admin. Every read endpoint of the public API has a tool; the ones left out on
purpose are listed below.
| Group | Tools | Scope |
|---|---|---|
| Segment | list_segments, get_segment, preview_segment, estimate_segment, list_segment_campaigns | segments:read |
| Campaign and message | list_campaigns, get_campaign, list_campaign_messages, get_campaign_stats, get_campaign_variant_stats, get_analytics_summary, list_templates, get_template, get_template_stats, list_template_versions, render_template | analytics:read |
| Delivery log | list_messages, get_message | analytics:read or notifications:write |
| Journey | list_journeys, get_journey, get_journey_stats, list_journey_runs, get_journey_run | analytics:read |
| User | get_app, search_users, get_user, get_user_journeys, get_user_preferences, list_categories | users:read |
| Analytics | get_organization_usage, list_event_definitions, get_event_definition, get_message_report, get_subscription_report, get_event_report, get_conversion_report, run_funnel, get_retention_report, get_engagement_heatmap, get_provider_report | analytics:read |
| In-app, web, inbox | list_in_app_messages, get_in_app_message, list_web_campaigns, get_web_campaign_report, list_forms, list_surveys, get_survey_report, get_inbox_metrics | analytics:read |
| Insights (RFM, predictions, decisioning) | get_rfm_overview, get_rfm_transitions, list_predictions, get_prediction, list_decision_agents, get_decision_agent_report, get_channel_score_overview | analytics:read |
| Flags, reports, paths, catalogs, geofences | list_feature_flags, get_feature_flag, run_path_analysis, run_report, list_saved_reports, get_saved_report, list_catalogs, get_recommendations, list_geofences | analytics:read |
| Scored users, Wallet | list_rfm_users, list_prediction_users, list_wallet_templates | users:read |
| Organization and apps | get_organization, list_organization_members, list_audit_logs, list_apps, list_channels, list_live_activity_types | admin |
| Users, subscriptions, companies | list_users, get_user_identity, get_user_inbox, get_subscription, get_subscription_identity, list_suppressions, list_companies, get_company, list_company_members, get_user_companies | users:read |
| Interests, per-person frequency, channel score | get_affinity_config, get_affinity_overview, get_channel_score_config, get_rfm_config | analytics:read |
| A person's interests and channel scores | get_user_affinity, list_channel_score_users, get_user_channel_scores | users:read |
| SQL queries and SQL segments | get_sql_schema, list_sql_runs, list_sql_segments | segments:read (the first two also with users:read) |
| Run a SQL query (SELECT only) | run_sql_query | users:read |
| Data pipeline | list_data_transforms, get_data_transform, list_data_sources, get_data_source, list_quarantined_records | users:read |
| Transformation preview (dry run) | preview_data_transform | admin |
| Warehouse segments | get_warehouse_segment_options, list_warehouse_segments, list_warehouse_segment_runs | segments:read |
| Warehouse connections, exports, reverse ETL, imports | list_warehouse_connections, get_warehouse_connection, list_warehouse_export_tables, list_warehouse_exports, get_warehouse_export, list_warehouse_export_runs, list_reverse_etl_syncs, get_reverse_etl_sync, list_reverse_etl_runs, list_imports, get_import | admin |
| Dynamic content | list_data_feeds, get_data_feed, list_content_tables, get_content_table, list_content_table_rows, list_promo_code_lists, get_promo_code_list, list_promo_codes | admin |
| Catalog, recommendation model, geofence, decision agent | get_catalog, list_catalog_items, get_catalog_item, get_recommendation_config, get_geofence, get_decision_agent | analytics:read |
| Wallet configuration | get_wallet_config, get_wallet_template | users:read |
| Webhooks | list_webhooks, get_webhook, list_webhook_deliveries | admin |
| Campaign and journey details | get_campaign_email_risk, list_campaign_deliveries, list_journey_versions | analytics:read |
| Message deliveries, WhatsApp templates | list_message_deliveries, list_whatsapp_templates | notifications:write |
| Pre-send check | run_campaign_preflight | notifications:write or mcp:write |
| Uninstall, Android delivery and control group reports | get_uninstall_report, get_android_delivery_report, get_global_control_report | analytics:read |
| Conversations, inbox, AI agent | list_conversations, get_conversation, list_conversation_messages, get_conversation_activity, list_canned_responses, list_inbox_teams, list_inbox_agents, list_ai_agents, get_ai_agent, list_ai_agent_runs, get_ai_agent_metrics | users:read |
| Custom AI agents | list_custom_agents, get_custom_agent, list_custom_agent_runs, list_custom_agent_drafts | analytics:read |
| Web personalization, web campaigns, forms, surveys, landing pages | get_web_campaign, list_web_personalizations, get_web_personalization, get_web_personalization_report, get_form, get_survey, list_landing_pages, get_landing_page, list_landing_page_versions | analytics:read |
| Form submissions, survey responses | list_form_submissions, list_survey_responses | users:read |
| Site search | get_site_search, list_site_search_sources, list_site_search_documents, get_site_search_report | analytics:read |
| Ad accounts and audiences | list_ad_accounts, list_ad_audiences, get_ad_audience, list_ad_audience_syncs | segments:read |
| Ad campaigns and reports | list_ad_campaigns, get_ad_campaign, get_ad_spend_report, get_ad_attribution_report, get_ad_ltv_report, list_conversions_api_configs, list_conversions_api_deliveries, list_ad_alert_rules, list_ad_alerts | analytics:read |
| Admin settings | list_sms_keywords, get_global_control, get_inbox_settings, get_email_deliverability, list_warmup_plans, get_warmup_plan, get_warmup_plan_usage, get_postmaster_reputation | admin |
| A person's timeline, events and messages; subscription list; recent events; conversation contact; assignment rules; channel summary | get_user_timeline, list_user_events, list_user_messages, list_subscriptions, list_recent_events, get_conversation_contact, list_assignment_rules, get_channel_status | users:read |
| Session volume, ad preview, pages and reporting settings | get_session_volume, get_ad_campaign_preview, list_ad_pages, get_ad_settings | analytics:read |
| Ad platforms | list_ad_platforms | segments:read |
| Tracing, logs, webhook details, list hygiene, integrations | trace_message, list_sdk_errors, list_request_logs, list_webhook_event_types, get_webhook_delivery, get_email_hygiene, get_email_hygiene_run, get_email_provider_insights, get_email_bounce_stats, list_integrations, get_integration, list_integration_runs | admin |
| Setup | get_setup_status, get_sdk_snippet | users:read |
| Setup, approval mode | create_app | mcp:write + permission to create apps (organization-wide key) |
| Setup, local stdio only | upload_channel_credentials | admin |
| Approval mode | create_campaign_draft, create_segment_draft, create_template_draft, create_journey_draft, create_in_app_draft, create_web_campaign_draft, create_web_personalization_draft, create_survey_draft, create_landing_page_draft, send_test_notification | mcp:write |
list_messages and get_message previously worked only with notifications:write; the read-only scope
analytics:read is now sufficient as well (existing keys keep working). get_organization_usage only works with an
organization-wide key. get_organization, list_organization_members, list_audit_logs, list_request_logs and
list_apps also need an organization-wide admin key. list_subscriptions returns only the last 8 characters of a
token.
run_sql_query runs a single SELECT only: the server validates the query, runs it in a read-only transaction with a
timeout and a row limit, and writes it to the query log; call get_sql_schema first for the views and limits.
preview_data_transform is a dry run and stores nothing. run_campaign_preflight produces the pre-send check report;
it does not change the campaign's content, audience or status, and does not approve or send it (only the latest report
summary is stored on the campaign). The mcp:write key that created the draft can also run the check.
Endpoints that are not offered as tools
- Credential management: API keys and SDK identity keys. Channel credentials are uploaded only on the local stdio
server, from a file path (
upload_channel_credentials); the remote server does not have this tool. - Responses that carry links: export jobs, the notification history CSV job, Wallet passes, the personal data export (data portability) and the ad creative list (signed image links).
- Zapier connection endpoints and the Zapier subscription list (target URLs).
- Endpoints that change something (other than the draft tools): sending, scheduling, publishing, status changes, deletion, running agents, site search crawls and document uploads, and creating links or secrets. Use the public API, the server SDKs or the dashboard for these.
Tools take an app_id parameter. With an app-bound key (or when PUSHIZO_APP_ID is set for stdio) you do not need to
provide it; with an organization-wide key it must be provided on every call.
Approval mode: write tools
- Draft tools create records but never send: a campaign always stays
draft, scheduling (send_at) is ignored and themcptag is added; segments are created as dynamic; templates get themcptag; journeys, in-app messages, web campaigns, web personalization experiences, surveys and landing pages are created asdraft(publishing and activation are not available on these endpoints). A web personalization draft needs theweb_personalizationplan feature. send_test_notificationis the only real send and it only reaches the app's test devices. Ifsubscription_idsare given, all of them must be test devices; otherwise403.- The tool descriptions ask the assistant to get the user's explicit approval before calling them (
destructiveHint=false). - The
mcp:writescope is enforced on the server side. Its dashboard equivalent is thecampaigns.create+segments.writepermissions; a member whose role lacks these cannot create a key containing this scope or grant it via OAuth. - Endpoints (public API, all
POST /v1/apps/{app_id}/mcp/…):campaign-drafts,segment-drafts,template-drafts,journey-drafts,in-app-drafts,web-campaign-drafts,web-personalization-drafts,survey-drafts,landing-page-drafts,test-sends. App creation:POST /v1/mcp/apps.
AI-assisted setup
When you say "set Pushizo up in my Flutter app", the assistant runs four steps. The contents of key files never enter the conversation, the model context or a tool result at any step.
| Step | Tool | What it does |
|---|---|---|
| 1. App | create_app | Creates the app; records the platforms and framework the same way as the dashboard setup wizard. |
| 2. Code | get_sdk_snippet | Returns the same setup code as the dashboard guide (with the app ID and API URL filled in). |
| 3. Key | upload_channel_credentials (local only) | Reads the key from a file path, uploads it and validates it with the provider. |
| 4. Verification | get_setup_status | Shows the channel statuses, whether the first subscriber has arrived and the missing steps. |
create_app
- Input:
name,platforms(ios,android,huawei,amazon,web,windows,email,sms), optionalframework(native,flutter,react-native,expo,kmp,unity,capacitor,cordova,maui) andweb_integration(code,wordpress). - Authorization: an organization-wide key, the
mcp:writescope and permission to create apps. Because the marketer role can also grantmcp:write, the scope alone is not enough: the key must have theadminscope, or the member who created the key (who gave consent, for OAuth) must currently be an owner or admin. An app-scoped key gets403. - The plan's app limit and the audit log entry are the same as creating an app from the dashboard. Setup continues where it left off on the Getting started page of the dashboard.
get_setup_status
Returns the selected platforms, the provider and status per channel, first_subscriber, sdk_platforms (which SDK
guides apply) and the missing list. Every missing step carries a dashboard link: /{org}/apps/{app}/channels for
credentials, /{org}/getting-started?app=… for the SDK and the first subscriber. No secrets or configuration details
(Key ID, bundle id, project id) are returned. A channel with local_upload: true can have its key uploaded from a file
on the local server or with the CLI.
get_sdk_snippet
Takes platform (27 platforms and languages), optional variant (code tab: Android kotlin/java, iOS
spm/cocoapods, script/npm/package for web frameworks) and lang (en or tr; the language of the comments
inside the code). The response has the install, init and extras blocks, produced by the same module as the
dashboard. For a native Android setup with Huawei or Amazon selected, the extra module lines come in android_modules.
Uploading the key from a file: upload_channel_credentials
Available only on the local stdio server; the remote (HTTP) server cannot see the files on your computer, so there
get_setup_status gives the dashboard link. The tool takes the path of the key; the server process on your
computer reads the file and sends it straight to the Pushizo API. It needs a key with the admin scope.
| Channel | Input |
|---|---|
apns | key_file (.p8) + team_id, bundle_id, environment; when key_id is omitted it is read from the AuthKey_XXXXXXXXXX.p8 file name. For a certificate, key_file (.p12) + optional p12_password_file or p12_password_env |
fcm | key_file (the Firebase service account .json; not google-services.json) |
hms | hms_app_id + client_secret_file or client_secret_env |
adm | client_id + client_secret_file or client_secret_env |
wns | package_sid + client_secret_file or client_secret_env |
webpush | subject (+ site_url); a VAPID key pair is generated unless you give one |
Security rules:
- Text secrets (
client_secret, the.p12password, the VAPID private key) are given only as a file path (*_file) or the name of an environment variable (*_env); no argument takes the value itself. - Files are read only from below the server's working directory; set
PUSHIZO_CREDENTIALS_DIRfor another directory. A path that leaves the directory (.., an absolute path, a symbolic link) is refused; the filesystem root cannot be the root directory. - Limits: 64 KB for a key file, 4 KB for a text secret. The format is checked locally (
.p8PKCS#8 PEM,.p12PKCS#12, service accounttype: service_account). - The result is status only:
saved,ok,status(active/invalid),error_code,hintand the file's name, size and the first 12 digits of its SHA-256 fingerprint. Contents and secrets are never written to the result or logs.
In Claude Desktop, put the keys in a folder and point the server at it:
{
"mcpServers": {
"pushizo": {
"command": "npx",
"args": ["-y", "@pushizo/mcp-server"],
"env": {
"PUSHIZO_API_KEY": "hs_live_…",
"PUSHIZO_CREDENTIALS_DIR": "/Users/you/pushizo-keys"
}
}
}
}In Claude Code and Cursor the server starts in the project directory; placing the key file below the project (outside git) is enough.
Example flow
You: Set Pushizo up in my Flutter app. iOS and Android. The APNs key is keys/AuthKey_ABCDE12345.p8,
the Firebase service account is keys/firebase.json. Team ID TEAM123456, bundle com.example.app.
Assistant: Shall I create the app "Store"? (approval) → create_app
Writing the setup code to pubspec.yaml and main.dart → get_sdk_snippet (platform: flutter)
Shall I upload the APNs key from the file? (approval) → upload_channel_credentials (apns, key_file)
Uploading the Firebase service account → upload_channel_credentials (fcm, key_file)
Open the app on a device; waiting for the first subscriber → get_setup_status
The first subscriber has arrived; setup is complete.Do not paste the contents of a key to the assistant; tell it the path. On the remote server (the Claude.ai connector) the third step is done from the dashboard or with the CLI:
pushizo channels apns --key-file keys/AuthKey_ABCDE12345.p8 --team-id TEAM123456 --bundle-id com.example.app --environment production
pushizo channels fcm --key-file keys/firebase.jsonLocal setup: Claude Desktop (stdio)
- Create a key in the dashboard under Developer → API keys. For read-only access use
segments:read+analytics:read; if you want drafts prepared, addmcp:writeas well. If possible, bind the key to a single app. - Edit the
claude_desktop_config.jsonfile (macOS:~/Library/Application Support/Claude/, Windows:%APPDATA%\Claude\; Claude Desktop → Settings → Developer → Edit Config):
{
"mcpServers": {
"pushizo": {
"command": "npx",
"args": ["-y", "@pushizo/mcp-server"],
"env": {
"PUSHIZO_API_KEY": "hs_live_…",
"PUSHIZO_API_URL": "https://api.pushizo.com",
"PUSHIZO_APP_ID": "app_…"
}
}
}
}- Restart Claude Desktop;
pushizoappears in the tool list.
Local setup: Cursor (stdio)
Cursor Settings → MCP (or .cursor/mcp.json in the project root):
{
"mcpServers": {
"pushizo": {
"command": "npx",
"args": ["-y", "@pushizo/mcp-server"],
"env": {
"PUSHIZO_API_KEY": "hs_live_…",
"PUSHIZO_API_URL": "https://api.pushizo.com",
"PUSHIZO_APP_ID": "app_…"
}
}
}
}For the remote server, use "url": "https://api.pushizo.com/mcp" instead of command in the same file; credentials are
provided with the Authorization: Bearer hs_live_… header or through OAuth discovery. Official listing in the Cursor and Claude
directories is a separate application made through an account; the contract in this repository is independent of that application.
| Environment variable | Required | Description |
|---|---|---|
PUSHIZO_API_KEY | yes | hs_live_… API key |
PUSHIZO_API_URL | no | API root URL (default https://api.pushizo.com). Must be https; for local development http://localhost:<port> or http://127.0.0.1:<port> is accepted |
PUSHIZO_APP_ID | no | App to use when a tool call does not provide app_id |
The server uses stdout only for MCP messages; diagnostics are written to stderr (visible in the Claude Desktop MCP
log). To run from the repository: npm run build --workspace @pushizo/mcp-server and
node packages/mcp-server/dist/stdio.js.
Remote server: Streamable HTTP
- URL:
POST {API_URL}/mcp(e.g.https://api.pushizo.com/mcp). - Stateless server, JSON responses (no session ID and no server-initiated SSE stream);
GETandDELETEreturn405. - Rate limit: 600 requests per minute per IP.
- Credentials:
Authorization: Bearer hs_live_…(a scoped API key or an access token obtained via OAuth). The key's IP allowlist is enforced (disallowed IP →403). If credentials are missing or invalid,401is returned together withWWW-Authenticate: Bearer resource_metadata="{API_URL}/.well-known/oauth-protected-resource/mcp"; MCP clients discover the OAuth flow from this header.
curl -s -X POST "$API/mcp" \
-H "Authorization: Bearer $HS_KEY" \
-H 'content-type: application/json' \
-H 'accept: application/json, text/event-stream' \
-d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'Claude.ai / Claude Desktop custom connector
Under Settings → Connectors → Add custom connector, enter {API_URL}/mcp as the URL. Claude opens the
Pushizo authorization screen in the browser (the OAuth flow below); once you approve, the connector works with the organization,
app and scopes you selected.
OAuth 2.1 authorization
Remote MCP clients can connect with browser-based OAuth 2.1 (authorization code + PKCE) without copying an API key.
Discovery
| Endpoint | Standard | Content |
|---|---|---|
GET /.well-known/oauth-protected-resource and …/oauth-protected-resource/mcp | RFC 9728 | resource ({API_URL}/mcp), authorization_servers: [API_URL], scopes_supported, bearer_methods_supported: ["header"], resource_documentation |
GET /.well-known/oauth-authorization-server | RFC 8414 | issuer (= API_URL), authorization_endpoint, token_endpoint, registration_endpoint, revocation_endpoint, supported methods |
Dynamic client registration: POST /oauth/register (RFC 7591)
- The default client type is public (
token_endpoint_auth_method: none); for confidential clientsclient_secret_postorclient_secret_basicare also supported. The client secret is returned only once, in the registration response. redirect_urisrules: anhttpsURL;httponly for loopback (localhost,127.0.0.1,[::1]); or a custom scheme containing a dot (e.g.com.pushizo.app:/cb).
curl -s -X POST "$API/oauth/register" -H 'content-type: application/json' -d '{
"client_name": "My MCP client",
"redirect_uris": ["http://127.0.0.1:33418/callback"]
}'Authorization: GET /oauth/authorize
Parameters: response_type=code, client_id, redirect_uri, code_challenge + code_challenge_method=S256
(PKCE is required, S256 only), state, optional scope (space-separated) and optional resource
(if given, it must be {API_URL}/mcp or {API_URL}). The admin scope cannot be granted via OAuth and is not advertised in the protected resource
metadata. If admin appears in a request together with other valid scopes, it is ignored; admin on its own
returns invalid_scope. The generic name mcp resolves to the read-only default (segments:read analytics:read). For automations that need full access, use a scoped API key. For an invalid client or an unregistered redirect_uri, no redirect happens
(400); other parameter errors are returned to the redirect_uri with error. For loopback addresses the port can be chosen by the
client (RFC 8252).
The browser is redirected to the consent screen in the dashboard (/oauth/consent); if there is no session, sign-in is requested first. On the
screen the user selects:
- the organization,
- optionally a single app (if none is selected, organization-wide),
- the scopes (only scopes their role can grant are selectable; default
segments:read analytics:read)
and approves. The browser then returns to the client's redirect_uri with code, state and iss (RFC 9207). The consent endpoint
is subject to the organization's SSO/MFA/IP policies and requires the apikeys.manage permission.
Token: POST /oauth/token
The body can be application/x-www-form-urlencoded or JSON.
grant_type=authorization_code+code,redirect_uri,client_id,code_verifier.grant_type=refresh_token+refresh_token,client_id.
curl -s -X POST "$API/oauth/token" \
-d grant_type=authorization_code -d code="$CODE" -d client_id="$CLIENT_ID" \
-d redirect_uri=http://127.0.0.1:33418/callback -d code_verifier="$VERIFIER"| Token | Lifetime | Note |
|---|---|---|
| Access token | 1 hour | It is a scoped API key (hs_live_…); used with Authorization: Bearer |
| Refresh token | 30 days | Replaced with a new one on every use (rotation). Reusing an old refresh token revokes the entire grant |
Revocation: POST /oauth/revoke (RFC 7009)
Revokes an access or refresh token.
Management and security
- Access granted via OAuth appears in the dashboard under Developer → API keys as
OAuth · <client name>; revoking it there ends the grant. - If the user leaves the organization or loses the relevant permissions, refresh fails; the client must obtain authorization again.
- Tokens, authorization codes and client secrets are stored only as SHA-256 hashes.
Scopes
| Scope | Tools it enables in MCP |
|---|---|
segments:read | Segment group (list_segments, get_segment, preview_segment, estimate_segment, list_segment_campaigns); SQL segment and query log, warehouse segment, ad account and ad audience read tools |
analytics:read | Campaign, template, journey, analytics, in-app, web growth and inbox read tools, plus list_messages / get_message; insight tools (RFM, predictions, decision agents, channel score, flags, reports, path analysis, catalogs and recommendations, geofences); interests, uninstall and Android delivery reports, custom AI agents, web personalization, form/survey/landing page details, site search, ad campaigns and reports |
notifications:write | list_messages, get_message (backward compatibility); list_message_deliveries, list_whatsapp_templates, run_campaign_preflight |
users:read | get_app, search_users, get_user, get_user_journeys, get_user_preferences, list_categories, list_rfm_users, list_prediction_users, list_wallet_templates; list_users, get_user_identity, get_user_inbox, get_subscription, get_subscription_identity, list_suppressions; companies, a person's interests and channel scores, data pipeline, conversations and inbox, AI agent, form submissions, survey responses, Wallet configuration and run_sql_query; setup: get_setup_status, get_sdk_snippet |
mcp:write | create_campaign_draft, create_segment_draft, create_template_draft, create_journey_draft, create_in_app_draft, create_web_campaign_draft, create_web_personalization_draft, create_survey_draft, create_landing_page_draft, send_test_notification; also run_campaign_preflight; create_app (also needs an organization-wide key and permission to create apps) |
admin | All; available with admin only: organization and app lists, channel configuration, warehouse connections and exports, reverse ETL, imports, dynamic content, webhooks, email deliverability and list hygiene, inbox settings, integration status, message tracing, SDK errors, request logs and preview_data_transform |
users:write and events:write are listed in the protected resource metadata, but no MCP tool uses them.
Follow the principle of least privilege: if the assistant only needs to read, do not grant mcp:write. User tools return personal
data; grant users:read to a key only when needed.
Error cases
- If a scope is missing, the tool returns an
isErrorresult (… (HTTP 403) — API anahtarının kapsamlarını kontrol edin., i.e. "check the API key's scopes"). app_id gerekli("app_id required") error: with an organization-wide key, passapp_idto the tool or setPUSHIZO_APP_ID.401(HTTP): the key/token is invalid, expired or revoked; an OAuth client should obtain a new access token with its refresh token.
Integrations: Segment, Amplitude, Mixpanel, GA4, PostHog, RudderStack, HubSpot, Salesforce, ActiveCampaign, Adobe Audience Manager, BlueConic, Median, Andromo, Retool, Zapier, Slack, WordPress, WooCommerce, Shopify, GTM
Ready-made integrations for connecting Pushizo to your data and automation stack without writing code (ADR-0031). Managed in the dashboard under Developer → Integrations (webhooks.manage permission: developer and admin roles).…
Data transformation and source pipeline
Fix, filter and enrich incoming event and person data before it is written, and bring in data from your own systems with webhooks or scheduled HTTP pulls. A faulty rule never drops data: the record is quarantined…