MCP connectors (reference)

TaTi does not embed all integration logic in the web binary: each domain (Slack, cloud, database…) is served by an MCP bridge — a small HTTP service compatible with the Model Context Protocol. Run them with Docker Compose (or manually), then register their URL in the UI: Settings → MCP servers (repo presets follow service names and /mcp paths). The UI preset grid is shown in Introduction — New MCP server.

Reminder

The ports below are default host ports from .env.example. You can change them: keep .env, docker compose, and the URL saved in TaTi consistent (e.g. if you change MCP_SLACK_PORT, http://localhost:…/mcp changes too).

MCP architecture (overview)

This diagram highlights the separation of responsibilities:

TaTi UI / API: orchestrates the conversation, decides which tools to call, and consolidates outputs.
MCP servers: expose specialized capabilities (Slack, GitHub, DB, cloud, observability) through a uniform MCP interface.
Target systems: remain behind MCP connectors, with their own permissions, tokens, and guardrails.
Main benefit: add a new business domain by plugging a new MCP server, without changing TaTi core.

MCP call flow (step by step)

Typical MCP call execution cycle:

The user sends a request in TaTi.
The model decides an MCP tool is needed (for example dashboard, SQL, ticket, observability).
TaTi calls the relevant MCP server with the useful context.
The MCP server executes the action on the external system and returns a structured result.
TaTi reformulates the final answer for the user using retrieved data.

Security note: effective permissions are those of the server token/account, not model permissions.

Summary table

Connector	Host port (default)	Typical URL (Compose network)	Key variables
OpenMetadata	8001	`http://mcp-openmetadata:8001/mcp`	`OPENMETADATA_URL`, `OPENMETADATA_JWT`
Metabase MCP	— (HTTPS)	`https://YOUR-METABASE/api/mcp`	OAuth 2.0 access token, `MB_SITE_URL` in Metabase
Tableau MCP	8024 → 3927 internal	`http://mcp-tableau:3927/tableau-mcp`	`TABLEAU_SERVER_URL`, `TABLEAU_SITE_NAME`, `TABLEAU_PAT_*` (or OAuth)
PostgreSQL	8002	`http://mcp-postgres:8002/mcp`	`MCP_POSTGRES_DATABASE_URL`, `MCP_POSTGRES_READ_ONLY`
PDF	8003	`http://mcp-pdf:8003/mcp`	`MCP_PDF_PUBLIC_BASE_URL`
Notion	8004	`http://mcp-notion:8004/mcp`	`MCP_NOTION_TOKEN`
GitHub	8007	`http://mcp-github:8007/mcp`	`MCP_GITHUB_TOKEN`
GitLab	8008	`http://mcp-gitlab:8008/mcp`	`MCP_GITLAB_TOKEN`, `MCP_GITLAB_URL`
Atlassian Rovo MCP	— (HTTPS)	`https://mcp.atlassian.com/v1/mcp`	OAuth 2.1 access token
monday.com MCP	— (HTTPS)	`https://mcp.monday.com/mcp`	OAuth access token (+ optional `Api-Version`)
Linear MCP	— (HTTPS)	`https://mcp.linear.app/mcp`	OAuth 2.1 access token
Elasticsearch	8009 → 8080 internal	`http://mcp-elasticsearch:8080/mcp`	`MCP_ELASTICSEARCH_*`
Discord	8010	`http://mcp-discord:8010/mcp`	`MCP_DISCORD_*`
Filesystem	8011	`http://mcp-filesystem:8011/mcp`	`MCP_FILESYSTEM_ROOT`
AWS	8012	`http://mcp-aws:8012/mcp`	`AWS_*`
Azure	8013	`http://mcp-azure:8013/mcp`	`AZURE_*`
GCP	8014	`http://mcp-gcp:8014/mcp`	`GCP_*`
Email SMTP	8015	`http://mcp-email:8015/mcp`	`SMTP_*`
Dagster	8016	`http://mcp-dagster:8016/mcp`	`DAGSTER_GRAPHQL_URL`, `DAGSTER_API_TOKEN`
Apache Airflow	8017	`http://mcp-airflow:8017/mcp`	`AIRFLOW_BASE_URL`, `AIRFLOW_USERNAME` / `AIRFLOW_PASSWORD`
dbt Cloud	8018	`http://mcp-dbt:8018/mcp`	`DBT_CLOUD_DISCOVERY_URL`, `DBT_CLOUD_TOKEN`, `DBT_CLOUD_ENVIRONMENT_ID`
dbt Core	8019	`http://mcp-dbt-core:8019/mcp`	`DBT_CORE_HOST_PROJECT`, `DBT_PROFILES_DIR`, `DBT_ALLOW_MUTATIONS`
Slack	8006	`http://mcp-slack:8006/mcp`	`MCP_SLACK_*`
Grafana	8020	`http://mcp-grafana:8020/mcp`	`MCP_GRAFANA_*`
Prometheus	8021	`http://mcp-prometheus:8021/mcp`	`MCP_PROMETHEUS_*`
n8n Builder MCP	8022	`http://mcp-n8n:3000/mcp`	`MCP_N8N_API_URL`, `MCP_N8N_API_KEY`, `MCP_N8N_AUTH_TOKEN`
MuleSoft DX MCP	8023	`http://mcp-mulesoft:8023/mcp`	`ANYPOINT_CLIENT_ID`, `ANYPOINT_CLIENT_SECRET`, `ANYPOINT_REGION`
Salesforce Hosted MCP	— (HTTPS)	`https://api.salesforce.com/...`	`MCP_SALESFORCE_MCP_URL`, `SALESFORCE_OAUTH_CLIENT_ID`, access token
HubSpot MCP	— (HTTPS)	`https://mcp.hubspot.com/`	HubSpot MCP auth app + OAuth access token
Zendesk MCP	8025	`http://mcp-zendesk:8025/mcp`	`ZENDESK_SUBDOMAIN`, `ZENDESK_EMAIL`, `ZENDESK_API_KEY`
ServiceNow MCP	8026	`http://mcp-servicenow:8026/mcp`	`SERVICENOW_INSTANCE_URL`, basic/token/OAuth auth
Datadog	— (HTTPS)	See Datadog section	`MCP_DATADOG_*` + headers
Brevo	— (HTTPS)	See Brevo section	`MCP_BREVO_MCP_URL`, `MCP_BREVO_MCP_TOKEN` (Bearer header)
Google Gmail / Calendar	— (HTTPS)	Google MCP endpoints	`GOOGLE_*`, OAuth
Moodle	— (HTTPS)	Moodle plugin URL	`MCP_MOODLE_*`

The sections below cover role, typical use, what the bridge usually exposes, configuration, and pitfalls.

MCP OpenMetadata

Role: MCP tools to query (and optionally change) metadata in OpenMetadata — tables, columns, glossaries, tags, lineage, business properties.

Typical use: “Which table feeds this report?”, “Who owns this dataset?”, “Show lineage around customers.”

Setup: Compose service mcp-openmetadata, port 8001. URL in TaTi: http://mcp-openmetadata:8001/mcp. Vars: OPENMETADATA_URL, OPENMETADATA_JWT, optional OPENMETADATA_ALLOW_MUTATIONS, OPENMETADATA_WRITE_CONFIRM_TOKEN.

WARNING

Catalog mutations are sensitive: keep OPENMETADATA_ALLOW_MUTATIONS=false until flows are validated.

MCP Metabase

Role: connect TaTi to the MCP server built into your Metabase instance (/api/mcp) to explore BI metadata, query tables and metrics, create questions, and build dashboards. Results are scoped to the permissions of the Metabase user who authorizes the client.

Typical use with TaTi

Search tables, metrics, and Metabase questions by keyword or natural language.
Construct and execute controlled queries through construct_query, execute_query, or query.
Create questions and dashboards from a business brief.
Prepare file-based dashboards using real table names, fields, and sample values.
Combine Metabase context with Slack, Linear, GitHub, Jira, or monday.com to explain a KPI.

Setup in Metabase

Use Metabase 0.61+ with AI/MCP features available.
Go to Admin → AI → MCP.
Enable MCP server.
Allow the needed clients (Cursor / VS Code, Claude, ChatGPT) or add your self-hosted client domain.
Make sure Site URL or MB_SITE_URL exactly matches the public URL used by TaTi.

Setup in TaTi

Settings → MCP Servers → preset Metabase.
Replace the URL with your endpoint: https://<your-metabase>/api/mcp.

Headers JSON:

json

{
  "Authorization": "Bearer <YOUR_METABASE_ACCESS_TOKEN>"
}

Enable the server, then Test.

WARNING

Metabase runs its own embedded OAuth 2.0 server. TaTi can call the endpoint with a Bearer token but does not run the full OAuth flow in the UI yet. If OAuth fails without a consent screen, check MB_SITE_URL / Site URL first.

Exposed tools

search, get_table, get_table_field_values, get_metric, get_metric_field_values, construct_query, execute_query, query, create_question, create_dashboard.

Official docs: Metabase MCP server

MCP Tableau

Role: connect TaTi to Tableau’s official MCP server (@tableau/mcp-server) to explore Tableau Cloud or Tableau Server: published data sources, workbooks, projects, views, custom views, VizQL Data Service queries, metadata, and Tableau Pulse.

Typical use with TaTi

List and search Tableau data sources, workbooks, projects, and views.
Read field metadata for a published data source.
Execute a VizQL query through query-datasource.
Retrieve CSV data or images from views and custom views.
Query Tableau Pulse: metric definitions, metrics, subscriptions, insight bundles, and insight briefs.
Combine Tableau KPIs with Metabase, Slack, Linear, Jira, GitHub, or monday.com context.

MCP architecture: Tableau + Atlassian

MCP architecture diagram connecting an MCP client, Tableau MCP, Atlassian MCP, Tableau Server/Cloud, and Confluence

In this model, the MCP client (Cursor, Claude Desktop, VS Code, or TaTi) orchestrates calls: the LLM decides which tools to use, then the client forwards requests to the right server. Tableau MCP turns requests into Tableau Server/Cloud API calls (REST, Metadata, VizQL, Pulse), while Atlassian MCP relays actions to Confluence/Jira. The LLM receives structured JSON/Markdown responses and combines them into the user-facing answer.

Deploying the Tableau MCP server

The official server is not a single SaaS endpoint: run @tableau/mcp-server and expose it with Streamable HTTP so TaTi can call it.

Typical configuration:

env

TRANSPORT=http
PORT=3927
SERVER=https://<your-tableau-cloud-or-server>
SITE_NAME=<site-content-url>
PAT_NAME=<personal-access-token-name>
PAT_VALUE=<token-value>

For multi-user deployments, Tableau recommends OAuth:

env

TRANSPORT=http
OAUTH_ISSUER=https://tableau-mcp.example.com

The MCP HTTP endpoint is usually:

txt

https://tableau-mcp.example.com/tableau-mcp

Compose + .env setup

Set .env: TABLEAU_SERVER_URL, TABLEAU_SITE_NAME, TABLEAU_PAT_NAME, TABLEAU_PAT_VALUE.
Start the service: docker compose up -d mcp-tableau.
The container runs @tableau/mcp-server with TRANSPORT=http on port 3927 (host 8024).

Setup in TaTi

Settings → MCP Servers → preset Tableau.
Default Compose URL: http://mcp-tableau:3927/tableau-mcp.
Headers JSON: leave empty in PAT mode (Tableau auth is handled by the mcp-tableau container).
Enable the server, then Test.

WARNING

Do not point TaTi directly at https://online.tableau.com or your Tableau Server URL. Point it at the Tableau MCP server, which then calls Tableau APIs (REST, Metadata, VDS, Pulse).

Exposed tools

list-datasources, list-workbooks, list-projects, list-views, list-custom-views, get-datasource-metadata, get-workbook, get-view-data, get-view-image, get-custom-view-data, get-custom-view-image, query-datasource, search-content, list-all-pulse-metric-definitions, generate-pulse-insight-brief.

Official docs: Tableau MCP · GitHub tableau/tableau-mcp

MCP PostgreSQL

Role: Run SQL against Postgres (often a business DB — may be the same instance as TaTi; scope carefully).

Typical use: schema exploration, read-only analytics (SELECT), avoid dumping copies manually.

Setup: Port 8002, URL http://mcp-postgres:8002/mcp. Vars: MCP_POSTGRES_DATABASE_URL, MCP_POSTGRES_READ_ONLY (true recommended for discovery).

TIP

Even read-only SELECT can load the DB — isolate networks and define SQL policy.

MCP PDF

Role: Generate PDFs and return download URLs.

Setup: Port 8003, MCP_PDF_PUBLIC_BASE_URL must be reachable by browsers following links.

MCP Notion

Role: Access Notion databases and pages via integration token permissions.

Setup: Port 8004, MCP_NOTION_TOKEN.

MCP Slack

Role: Read history (per scopes) and post via a bot; optionally restrict channels.

Setup: Port 8006. Vars: MCP_SLACK_BOT_TOKEN, MCP_SLACK_TEAM_ID, MCP_SLACK_CHANNEL_IDS (recommended to limit channels).

MCP Discord

Role: Discord equivalent of the Slack bridge — guild channels and bot permissions.

Setup: Port 8010, MCP_DISCORD_BOT_TOKEN, MCP_DISCORD_GUILD_ID, MCP_DISCORD_CHANNEL_IDS.

MCP Filesystem

Role: List/read (and sometimes write) files under a single root (FILESYSTEM_ROOT).

Setup: Port 8011, MCP_FILESYSTEM_ROOT. In production use a dedicated volume and strict permissions.

MCP GitHub

Role: Automate GitHub workflows — issues, PRs, CI context.

Setup: Port 8007, MCP_GITHUB_TOKEN, MCP_WRITE_CONFIRM_TOKEN for guarded writes.

MCP GitLab

Role: Projects, issues, merge requests — SaaS or self-hosted GitLab.

Setup: Port 8008, MCP_GITLAB_TOKEN, MCP_GITLAB_URL.

MCP Elasticsearch

Role: Full-text search and index operations — depends on MCP image and stack version.

Setup: Host 8009 mapped to 8080 in container — URL http://mcp-elasticsearch:8080/mcp. Vars: MCP_ELASTICSEARCH_URL, credentials, optional MCP_ELASTICSEARCH_SSL_SKIP_VERIFY.

MCP AWS

Role: AWS APIs — often read-only inventory; mutations are high-risk.

Setup: Port 8012, AWS_REGION, AWS_PROFILE or static keys / session token.

MCP Azure

Role: Azure resources via ARM or APIs exposed by the bridge.

Setup: Port 8013, Service Principal or AZURE_ACCESS_TOKEN. Prefer Reader for exploration.

MCP GCP

Role: GCP project resources — Compute, GKE, storage, IAM reads per service account.

Setup: Port 8014, GCP_PROJECT_ID, GCP_REGION, GCP_SERVICE_ACCOUNT_JSON (never commit).

MCP Email (SMTP)

Role: Send mail via SMTP — no IMAP inbox in scope.

Setup: Port 8015, SMTP_*, SMTP_ALLOWED_RECIPIENTS to prevent misuse.

MCP Dagster

Role: Dagster via GraphQL — assets, jobs, runs, partitions.

Setup: Port 8016, DAGSTER_GRAPHQL_URL, DAGSTER_API_TOKEN, DAGSTER_ALLOW_MUTATIONS.

MCP Apache Airflow

Role: Query and control Apache Airflow via the stable REST API (/api/v1): DAGs, dag runs, tasks; trigger or pause when mutations are enabled.

Typical use

“List active DAGs”, “last run status for daily_etl”, “trigger a run with this conf” (if mutations are on).

Bridge capabilities

Read: list DAGs, DAG detail, list/detail dag runs, list tasks for a DAG.
Write (optional): trigger a dag run, pause/unpause DAG — only if AIRFLOW_ALLOW_MUTATIONS=true and the Airflow account allows it.

Configuration

Port: 8017 (MCP_AIRFLOW_PORT).
TaTi URL: http://mcp-airflow:8017/mcp (Compose) or http://localhost:8017/mcp from the host.
Variables: AIRFLOW_BASE_URL (webserver root URL without /api/v1), AIRFLOW_USERNAME / AIRFLOW_PASSWORD (Basic auth), optional AIRFLOW_API_TOKEN (Bearer), AIRFLOW_SSL_VERIFY, AIRFLOW_ALLOW_MUTATIONS.

MCP dbt Cloud (Discovery API)

Role: Query dbt Cloud metadata via the Discovery GraphQL API — models, sources, custom lineage queries.

Setup: Port 8018 (MCP_DBT_PORT). TaTi URL: http://mcp-dbt:8018/mcp.

Variables: DBT_CLOUD_DISCOVERY_URL (from Account settings → Access URLs), DBT_CLOUD_TOKEN (service token with Discovery access), optional DBT_CLOUD_ENVIRONMENT_ID for shortcut tools, DBT_SSL_VERIFY.

Scope: This bridge targets dbt Cloud Discovery. For a local dbt Core project, use MCP dbt Core.

MCP dbt Core (local CLI)

Role: Run dbt Core CLI in a container against a mounted project directory — parse, ls, compile, manifest summary, and optionally deps / run / test / build when mutations are enabled.

Setup: Port 8019 (MCP_DBT_CORE_PORT). TaTi URL: http://mcp-dbt-core:8019/mcp.

Variables: DBT_CORE_HOST_PROJECT (host path mounted at /workspace), DBT_CORE_PROJECT_DIR (container path, default /workspace), optional DBT_PROFILES_DIR, DBT_CORE_TARGET (--target), DBT_ALLOW_MUTATIONS (false by default — blocks deps/run/test/build).

Notes: Image includes dbt-postgres; extend the Dockerfile for other adapters. Warehouse credentials are required for compile/run against a live warehouse.

MCP Grafana

Role: Official Grafana MCP — dashboards, folders, alerts.

Setup: Port 8020, MCP_GRAFANA_URL, MCP_GRAFANA_SERVICE_ACCOUNT_TOKEN, optional org ID.

MCP Prometheus

Role: PromQL queries, targets, series metadata.

Setup: Port 8021, MCP_PROMETHEUS_URL, optional auth, MCP_PROMETHEUS_SSL_VERIFY.

MCP Datadog

Role: Hosted Datadog MCP (not a default Compose container) — APM, logs, metrics per product.

Setup: MCP_DATADOG_MCP_URL (US/EU). Keys often passed as headers in TaTi (DD_API_KEY, DD_APPLICATION_KEY).

MCP Brevo

Role: Brevo-hosted MCP (no Docker container in compose) — contacts, lists, transactional email, campaigns, templates, and analytics depending on your account modules.

Typical use with TaTi

“List my contacts”, “send a transactional email to …”, “create a campaign”, “summarize send stats”.
Marketing / CRM automation from chat without opening the Brevo UI.

TaTi setup

Generate an MCP token: Brevo → Account → SMTP & API → API Keys → enable Create MCP server API key.
Settings → MCP servers → Brevo preset (or manual entry):
- URL: https://mcp.brevo.com/v1/brevo/mcp
- Headers JSON:
  json
```
{
  "Authorization": "Bearer <YOUR_BREVO_MCP_TOKEN>"
}
```
Enable the server and run Test in the UI.
In chat: short prompts like “use the Brevo server to …”.

.env reference (optional)

bash

MCP_BREVO_MCP_URL=https://mcp.brevo.com/v1/brevo/mcp
MCP_BREVO_MCP_TOKEN=

Secrets stay in TaTi MCP headers (like Datadog), not in the browser.

Common pitfalls

Mixing a classic API v3 key with the MCP token — this MCP needs the dedicated MCP key.
Do not pass the token in the URL; use Authorization: Bearer … header.
Early access: tool list may vary by Brevo account.

Official docs: Brevo MCP Protocol

MCP HubSpot

Role: connect TaTi to the official remote HubSpot MCP server for CRM and marketing data: contacts, companies, deals, tickets, activities, campaigns, marketing content, and properties.

This server is separate from HubSpot’s local Developer MCP server. The Developer MCP is for HubSpot developer tooling through the CLI; the remote https://mcp.hubspot.com/ server is the right choice for CRM / marketing data in TaTi.

Typical use with TaTi

Search contacts, companies, deals, tickets, or owners from chat.
Summarize sales pipeline, recent activities, or marketing campaign performance.
Create or update contacts, deals, tickets, notes, tasks, or products depending on granted permissions.

HubSpot tools

get_user_details
search_crm_objects
get_crm_objects
manage_crm_objects
search_properties
get_properties
search_owners
get_campaign_contacts_by_type
get_campaign_analytics
get_campaign_asset_types
get_campaign_asset_metrics
submit_feedback

HubSpot prerequisites

Create an MCP auth app in HubSpot: Development → MCP Auth Apps → Create MCP auth app.
Configure a Redirect URL compatible with the OAuth client you use.
Copy the Client ID and Client secret.
Run an OAuth flow with PKCE to obtain an access token. HubSpot requires PKCE.
Pass that token to TaTi in the MCP server headers.

TaTi setup

Settings → MCP servers → HubSpot preset.
URL: https://mcp.hubspot.com/.

Headers JSON:

json

{
  "Authorization": "Bearer <YOUR_HUBSPOT_ACCESS_TOKEN>"
}

Enable the server and run Test.

WARNING

TaTi can call the remote MCP server with a Bearer token, but it does not yet run HubSpot’s OAuth PKCE flow inside the UI. Generate the token with a PKCE-capable client such as MCP Inspector or a dedicated OAuth integration.

TIP

Effective permissions follow the HubSpot user who authorizes the app. If HubSpot Sensitive Data is enabled, activity objects such as calls, emails, meetings, notes, and tasks may be blocked by the MCP server.

Official docs: HubSpot MCP server

MCP Zendesk

Role: connect TaTi to Zendesk Support through reminia/zendesk-mcp-server, an open-source MCP server. The upstream server runs over stdio, so TaTi uses the local Docker bridge mcp-zendesk to expose an HTTP /mcp endpoint compatible with the app.

Typical use with TaTi

Analyze ticket trends: volume, recurring topics, friction points, escalations.
Summarize conversations and suggest tags, categories, or triage actions.
Read recent tickets, retrieve a specific ticket, and inspect ticket comments.
Create tickets, add comments, and update status, priority, assignee, tags, or custom fields.
Identify Help Center documentation gaps from repeated customer questions.

Exposed tools

get_tickets: list latest tickets with pagination and sorting.
get_ticket: retrieve a ticket by ID.
get_ticket_comments: retrieve comments for a ticket.
create_ticket_comment: add a public or private ticket comment.
create_ticket: create a ticket with subject, description, priority, type, tags, and custom fields.
update_ticket: update status, priority, type, assignee, requester, tags, custom fields, or due date.

The server also exposes the zendesk://knowledge-base resource for Help Center articles, plus the analyze-ticket and draft-ticket-response prompts.

Setup in Zendesk

Create or choose a technical Zendesk user.
Generate a Zendesk API key.
Fill .env: ZENDESK_SUBDOMAIN, ZENDESK_EMAIL, ZENDESK_API_KEY.

Setup in TaTi

Start or restart the mcp-zendesk container with Docker Compose.
Settings → MCP Servers → preset Zendesk.
URL: http://mcp-zendesk:8025/mcp.
Headers JSON: leave empty; Zendesk authentication is handled by the container through .env.
Enable the server, then Test.

TIP

This server can write to Zendesk (create_ticket, create_ticket_comment, update_ticket). Use a technical account with limited permissions if you want to reduce scope.

Docs / resources: reminia/zendesk-mcp-server

MCP ServiceNow

Role: connect TaTi to ServiceNow through michaelbuckner/servicenow-mcp, an open-source MCP server. The upstream server runs over stdio, so TaTi uses the local Docker bridge mcp-servicenow to expose an HTTP /mcp endpoint compatible with the app.

Typical use with TaTi

Search incidents, users, Knowledge articles, and ServiceNow records.
Read available tables, schemas, and table records.
Create or update incidents.
Add customer-visible comments or internal work notes.
Run natural language searches and updates.
Update ServiceNow scripts (script includes, business rules, client scripts) carefully.

Exposed resources

servicenow://incidents and servicenow://incidents/{number}
servicenow://users
servicenow://knowledge
servicenow://tables and servicenow://tables/{table}
servicenow://schema/{table}

Exposed tools

create_incident
update_incident
search_records
get_record
perform_query
add_comment
add_work_notes
natural_language_search
natural_language_update
update_script

ServiceNow setup

Create or choose a technical account with API access.
Limit roles to the expected scope (incidents, knowledge, tables/scripts only if needed).
Choose authentication: basic auth, token, or OAuth client credentials + username/password.

TaTi setup

Fill .env:

bash

SERVICENOW_INSTANCE_URL=https://your-instance.service-now.com/
SERVICENOW_USERNAME=
SERVICENOW_PASSWORD=
# Alternative: SERVICENOW_TOKEN or SERVICENOW_CLIENT_ID/SECRET + username/password
SERVICENOW_TOKEN=
SERVICENOW_CLIENT_ID=
SERVICENOW_CLIENT_SECRET=
MCP_SERVICENOW_PORT=8026

Start or restart the mcp-servicenow container.
Settings → MCP Servers → preset ServiceNow.
URL: http://mcp-servicenow:8026/mcp.
Headers JSON: leave empty; ServiceNow authentication is handled by the container through .env.
Enable the server, then Test.

WARNING

This server can modify ServiceNow (create_incident, update_incident, add_comment, add_work_notes, natural_language_update, update_script). Use a restricted technical account and avoid script permissions in production without guardrails.

Docs / resources: michaelbuckner/servicenow-mcp · ServiceNow MCP Client

MCP Atlassian / Jira + Confluence

Role: connect TaTi to the Atlassian Rovo remote MCP server for Jira, Confluence, and Compass through one connector. This is the right integration for combining delivery status, tickets, docs, and product context.

Typical use with TaTi

Cross-check Jira issues, GitHub PRs, and Confluence pages in one prompt.
Summarize Jira issues with their Confluence context.
Create or enrich Jira issues and Confluence pages depending on permissions.
Find issues without PRs, PRs without issues, or features without documentation.

TaTi setup

Settings → MCP servers → Atlassian (Jira + Confluence) preset.
URL: https://mcp.atlassian.com/v1/mcp.

Headers JSON:

json

{
  "Authorization": "Bearer <YOUR_ATLASSIAN_ACCESS_TOKEN>"
}

Enable the server and run Test.

WARNING

The Atlassian server uses OAuth 2.1 and respects the Atlassian user’s permissions. TaTi can call the endpoint with a Bearer token, but it does not yet run the full OAuth flow inside the UI.

TIP

Atlassian recommends the /v1/mcp endpoint. Avoid the legacy SSE endpoint (/v1/sse), which is being deprecated.

Official docs: Atlassian Rovo MCP Server

MCP monday.com

Role: connect TaTi to the hosted monday.com MCP server for workspace data: boards, items, columns, updates, monday docs, dashboards, and insights. Use it for project management, monday CRM, and operations workflows alongside or instead of Jira/Atlassian.

Typical use with TaTi

Project reporting: sprint summaries, team workload, deadlines across boards (including monday dev).
Turn meeting notes into structured items with owners, priority, and due dates.
Cross-team visibility: “What’s blocking the launch?” across Product, Marketing, and RevOps boards.
Ops / support: incidents, requests, and checklists with owners and priorities.
monday CRM: leads, deals, pipeline stages, and next steps from call notes.
Documentation: specs, SOPs, and retros as monday docs in a workspace or attached to items.

Setup in TaTi

Settings → MCP Servers → preset monday.com.
URL: https://mcp.monday.com/mcp.

Headers JSON:

json

{
  "Authorization": "Bearer <YOUR_MONDAY_ACCESS_TOKEN>"
}

Optional — monday API version:

json

{
  "Authorization": "Bearer <YOUR_MONDAY_ACCESS_TOKEN>",
  "Api-Version": "2026-07"
}

Enable the server, then Test.

WARNING

monday MCP uses OAuth (consent on first connect). If connection fails, ask your admin to check Admin → Permissions → AI Connectors. TaTi can call the endpoint with a Bearer token but does not run the full OAuth flow in the UI yet.

TIP

For delivery cross-analysis, enable monday.com + GitHub + optionally Atlassian or HubSpot in one prompt by naming each integration.

Official docs: Get started with monday MCP · Platform MCP overview

MCP Linear

Role: connect TaTi to the Linear remote MCP server for issues, projects, comments, and cycles. Use it to combine product roadmap, bugs, priorities, and execution tracking with GitHub, Slack, Jira, or monday.com.

Typical use with TaTi

Search, summarize, and prioritize Linear issues.
Create or update issues, projects, and comments according to OAuth permissions.
Combine Linear issues, GitHub PRs, and Slack discussions in one prompt.
Identify bugs without owners, risky projects, or cycle blockers.
Prepare product rollups: roadmap, project progress, triage, and risks.

Setup in TaTi

Settings → MCP Servers → preset Linear.
URL: https://mcp.linear.app/mcp.

Headers JSON:

json

{
  "Authorization": "Bearer <YOUR_LINEAR_ACCESS_TOKEN>"
}

Enable the server, then Test.

WARNING

Linear uses an authenticated remote MCP with Streamable HTTP transport and OAuth 2.1. TaTi can call the endpoint with a Bearer token but does not run the full OAuth flow in the UI yet.

TIP

For clients that do not support native remote MCP yet, Linear recommends mcp-remote. In TaTi, use the preset HTTP URL directly.

Official docs: Linear MCP server

MCP Salesforce

Role: connect TaTi to Salesforce hosted MCP servers. They expose MCP-ready Salesforce products without a local Docker bridge: SObjects, Flow, Apex, API Catalog, Data 360, Prompt Builder, and Tableau Next depending on your org configuration.

Typical use with TaTi

Read and query standard or custom Salesforce objects through platform/sobject-reads.
Run controlled automations through Flow or Apex methods exposed as MCP tools.
Query Data 360 or Tableau Next for CRM / analytics questions in natural language.

Salesforce prerequisites

Create an External Client App in Salesforce. Classic Connected Apps are not supported for Salesforce MCP clients.
Enable OAuth with Access MCP servers (mcp_api) and refresh_token scopes.
Select JWT-based access tokens for named users and Require PKCE.
Copy the Consumer Key (SALESFORCE_OAUTH_CLIENT_ID).
Obtain an OAuth access token for the authorized user, then pass it to TaTi as Authorization: Bearer ....

MCP URLs

Production: https://api.salesforce.com/platform/mcp/v1/<SERVER-NAME>
Sandbox / scratch org: https://api.salesforce.com/platform/mcp/v1/sandbox/<SERVER-NAME>

Useful server examples:

platform/sobject-reads: SObject reads and queries, recommended first.
platform/sobject-all: full CRUD, use only in controlled environments.
platform/sobject-mutations: create and update.
platform/sobject-deletes: delete operations only.
analytics/tableau-next: Tableau Next / analytics agent.

TaTi setup

Settings → MCP servers → Salesforce preset.
Keep the default read-only URL to start: https://api.salesforce.com/platform/mcp/v1/platform/sobject-reads
For sandbox, replace it with: https://api.salesforce.com/platform/mcp/v1/sandbox/platform/sobject-reads

Replace the header:

json

{
  "Authorization": "Bearer <YOUR_SALESFORCE_ACCESS_TOKEN>"
}

Enable the server and run Test.

WARNING

Start with platform/sobject-reads. sobject-all, mutations, and deletes expose write/delete capabilities and should be limited to explicitly authorized users and environments.

.env reference (optional)

bash

MCP_SALESFORCE_MCP_URL=https://api.salesforce.com/platform/mcp/v1/platform/sobject-reads
SALESFORCE_OAUTH_CLIENT_ID=
SALESFORCE_ACCESS_TOKEN=

MCP n8n (Builder + Instance)

Goal: let TaTi create, edit, list, and run n8n workflows from chat.

Choose the right mode

n8n Builder MCP (recommended for build/edit):
- Local Docker service mcp-n8n (ghcr.io/talent-tally/tati-mcp-n8n:latest).
- TaTi URL: http://mcp-n8n:3000/mcp.
- Exposes workflow CRUD/build tools when N8N_MODE=true and API access is valid.
n8n Instance MCP (native n8n endpoint):
- Direct n8n endpoint: https://<your-domain>/mcp-server/http.
- Auth with n8n MCP token (Authorization: Bearer <token>).
- Good for native instance-level access; depending on n8n version/config, tool coverage may be more run/ops-oriented.

n8n-side prerequisites

Enable Instance-level MCP in n8n: Settings -> Instance-level MCP.
Enable target workflows as Available in MCP.
Generate the required secrets:
- n8n API Key (for Builder MCP through n8n API).
- MCP Access Token (for native Instance MCP).

`.env` variables (local Builder MCP)

bash

MCP_N8N_API_URL=https://your-subdomain.app.n8n.cloud
MCP_N8N_API_KEY=...
MCP_N8N_AUTH_TOKEN=change-me-n8n-mcp-token
MCP_N8N_WEBHOOK_SECURITY_MODE=strict
MCP_N8N_PORT=8022
MCP_N8N_IMAGE=ghcr.io/talent-tally/tati-mcp-n8n:latest

Important

MCP_N8N_API_URL must be the base instance URL (for example https://xxx.app.n8n.cloud), without /api/v1.
Keep token/key values literal in .env (special characters are supported).

Start command (exact)

From repo root:

bash

docker compose up -d mcp-n8n app

Then verify:

bash

docker compose logs --tail=100 mcp-n8n
curl -s http://localhost:8022/health

Expected:

logs showing n8n MCP ... running on 0.0.0.0:3000
health endpoint returns OK.

TaTi UI setup

In Settings -> MCP servers:

preset n8n (Builder MCP):
- URL: http://mcp-n8n:3000/mcp
- Headers JSON:
  json
```
{
  "Authorization": "Bearer <MCP_N8N_AUTH_TOKEN>"
}
```
optional preset n8n (Instance MCP):
- URL: https://<your-subdomain>.app.n8n.cloud/mcp-server/http
- Header Authorization: Bearer <YOUR_N8N_MCP_TOKEN>

Recommended end-to-end smoke test

In TaTi, enable only n8n Builder.
Send short prompts:
- List my n8n workflows
- then Create a simple workflow with a trigger and email send
Confirm workflow creation/changes in n8n UI.

Common failures and fixes

Missing tools (n8n_create_workflow, n8n_list_workflows)
- check N8N_MODE=true on mcp-n8n
- check MCP_N8N_API_URL does not include /api/v1
- verify MCP_N8N_API_KEY validity.
“Cannot test via API - Manual Trigger...”
- expected behavior: Manual Trigger does not support external API test; use webhook/schedule/chat trigger for API-driven testing.
Port conflict concerns
- host uses 8022; 3000 is internal container port.
Agent gives incomplete answers
- verify TaTi is using the intended MCP server (Builder vs Instance), then retry with shorter prompts.

MCP MuleSoft DX

Role: connect TaTi to Anypoint Platform through the official MuleSoft DX MCP Server. MuleSoft’s server runs over stdio (npx mulesoft-mcp-server start); TaTi runs it through the local mcp-mulesoft bridge, which exposes an HTTP /mcp endpoint for the app.

Use cases with TaTi

List deployed Mule applications and their runtime status.
Search / inspect Exchange assets and API specs depending on granted scopes.
Query Runtime Manager, Monitoring, API Manager, and available platform metrics.
Prepare incident diagnostics or environment reviews directly from chat.

MuleSoft prerequisites

Create an Anypoint Connected App that acts on its own behalf.
Copy the Client ID and Client Secret.
Grant the scopes needed for your use cases: Read Applications, View Organization, Monitoring Viewer, View APIs Configuration, Exchange Viewer, etc.
Set ANYPOINT_REGION if needed: PROD_US, PROD_EU, PROD_CA, or PROD_JP.

Configuration

env

ANYPOINT_CLIENT_ID=your-client-id
ANYPOINT_CLIENT_SECRET=your-client-secret
ANYPOINT_REGION=PROD_US
MCP_MULESOFT_PORT=8023

Compose: mcp-mulesoft, port 8023.
TaTi URL: http://mcp-mulesoft:8023/mcp.
UI: Settings → MCP servers → MuleSoft preset → Test.

WARNING

Effective permissions are the Connected App permissions. Start with read-only scopes until mutation / deployment flows are explicitly validated.

Google Gmail & Calendar (remote MCP)

Role: Gmail and Calendar via Google-hosted MCP endpoints.

Setup: MCP_GMAIL_MCP_URL, MCP_GOOGLE_CALENDAR_MCP_URL, GOOGLE_CLOUD_PROJECT_ID, OAuth vars. Configure headers in TaTi if required.

MCP Moodle

Role: Moodle plugin MCP endpoint for courses and activities.

Setup: MCP_MOODLE_MCP_URL, MCP_MOODLE_TOKEN — least-privilege token.

Deprecations and alternatives

Elasticsearch MCP image: .env.example may note Elastic-side evolution — watch release notes for recommended endpoints.

MCP connectors (reference) ​

MCP architecture (overview) ​

MCP call flow (step by step) ​

Summary table ​

MCP OpenMetadata ​

MCP Metabase ​

MCP Tableau ​

MCP PostgreSQL ​

MCP PDF ​

MCP Notion ​

MCP Slack ​

MCP Discord ​

MCP Filesystem ​

MCP GitHub ​

MCP GitLab ​

MCP Elasticsearch ​

MCP AWS ​

MCP Azure ​

MCP GCP ​

MCP Email (SMTP) ​

MCP Dagster ​

MCP Apache Airflow ​

MCP dbt Cloud (Discovery API) ​

MCP dbt Core (local CLI) ​

MCP Grafana ​

MCP Prometheus ​

MCP Datadog ​

MCP Brevo ​

MCP HubSpot ​

MCP Zendesk ​

MCP ServiceNow ​

MCP Atlassian / Jira + Confluence ​

MCP monday.com ​

MCP Linear ​

MCP Salesforce ​

MCP n8n (Builder + Instance) ​

Choose the right mode ​

n8n-side prerequisites ​

.env variables (local Builder MCP) ​

Start command (exact) ​

TaTi UI setup ​

Recommended end-to-end smoke test ​

Common failures and fixes ​

MCP MuleSoft DX ​

Google Gmail & Calendar (remote MCP) ​

MCP Moodle ​

Deprecations and alternatives ​

See also ​

MCP connectors (reference)

MCP architecture (overview)

MCP call flow (step by step)

Summary table

MCP OpenMetadata

MCP Metabase

MCP Tableau

MCP PostgreSQL

MCP PDF

MCP Notion

MCP Slack

MCP Discord

MCP Filesystem

MCP GitHub

MCP GitLab

MCP Elasticsearch

MCP AWS

MCP Azure

MCP GCP

MCP Email (SMTP)

MCP Dagster

MCP Apache Airflow

MCP dbt Cloud (Discovery API)

MCP dbt Core (local CLI)

MCP Grafana

MCP Prometheus

MCP Datadog

MCP Brevo

MCP HubSpot

MCP Zendesk

MCP ServiceNow

MCP Atlassian / Jira + Confluence

MCP monday.com

MCP Linear

MCP Salesforce

MCP n8n (Builder + Instance)

Choose the right mode

n8n-side prerequisites

`.env` variables (local Builder MCP)

Start command (exact)

TaTi UI setup

Recommended end-to-end smoke test

Common failures and fixes

MCP MuleSoft DX

Google Gmail & Calendar (remote MCP)

MCP Moodle

Deprecations and alternatives

See also