How do I install Data360?

Copy the stdio config for Cursor, Claude Desktop, Claude Code, or VS Code on this page. It uses the publisher's install, not a PlugMCP proxy.

Does Data360 need an API key?

No required environment variable is listed on this record.

What transport does Data360 use?

Data360 runs locally over stdio.

When was Data360 last checked?

Last checked 2026-10-28. A Model Context Protocol (MCP) server for Salesforce Data Cloud — 60+ tools for SQL queries, data streams, segments, calculated insights, and more.

Data360 MCP 伺服器

:zap: FLASH(Framework Layer for Agentic Serverless Headless Data360)

:ai-bot:A Headless, API-Native & Agentic AI Framework for Salesforce Data360

FLASH is a headless, API-native, agentic AI framework that transforms how Salesforce Data360 Implementations are delivered and operated. Built on the custom Model Context Protocol (MCP), FLASH exposes 60+ intelligent APIs and operational tools that allow AI agents to autonomously execute Data360 workflows through natural language — eliminating dependency on manual UI-driven configuration. From ingestion and modeling to segmentation, querying, and insights generation, FLASH enables organizations to operate Salesforce Data360 through a scalable headless and agentic architecture, accessible from:

:computer: Cursor
:desktop_computer: Claude Desktop
:gear: Agentforce
:ai-bot:Custom AI Agents

FLASH creates a reusable AI execution layer that standardizes, accelerates, and simplifies Data 360 operations across the enterprise.

Prerequisites
Step-by-Step Setup
Verify It Works
Tool Catalog
Configuration Reference
Architecture
Security
Troubleshooting

Prerequisites

Before starting, make sure you have:

Requirement	How to check	Install if missing
Python 3.10 or newer (3.13 recommended)	`python3 --version`	python.org/downloads
pip (Python package manager)	`pip3 --version`	Comes with Python; or `python3 -m ensurepip`
git	`git --version`	git-scm.com
A Salesforce org with Data Cloud enabled	Setup > Data Cloud	Contact your Salesforce admin
A Salesforce user with Data Cloud permissions	—	Use the narrowest permission set that covers the enabled tools. Keep writes and deletes disabled unless that admin access is required.

Step-by-Step Setup

Follow these steps in order. The entire setup takes about 15 minutes.

Step 1: Clone the Repository

git clone https://github.com/rishiganesh25/data360-mcp.git
cd data360-mcp

Step 2: Create a Python Virtual Environment

python3 -m venv .venv

Activate it:

# macOS / Linux
source .venv/bin/activate

# Windows (PowerShell)
.\.venv\Scripts\Activate.ps1

# Windows (cmd)
.\.venv\Scripts\activate.bat

You should see (.venv) in your terminal prompt.

Step 3: Install Dependencies

pip install -r requirements.txt

This installs the MCP SDK, requests, pydantic, and other required packages.

Step 4: Install the Salesforce CLI and log in

The server does not use a Connected App consumer secret. It reads the access token from an authenticated Salesforce CLI org, and it will not start unless the sf command is installed.

npm install -g @salesforce/cli
sf org login web --alias data360

Use --instance-url https://test.salesforce.com for a sandbox, or your My Domain URL for a production org. The user must have Data Cloud permissions.

Step 5: Verify the Server Can Start

python server.py

You should see Starting MCP server in the output. Press Ctrl+C to stop it.

Step 6: Point the server at that org

cp .env.example .env
chmod 600 .env

Set the alias you used above:

SF_TARGET_ORG=data360

Leave it unset to use the CLI's default target org. Do not put a consumer key or consumer secret in this file.

Step 7: Wire It Into Your MCP Client

Pick the MCP client you're using and follow the config below.

In all examples, replace /absolute/path/to/data360-mcp with the actual full path where you cloned the repo (e.g., /Users/yourname/projects/data360-mcp).

If you used a virtual environment (Step 2), replace python with .venv/bin/python in the command.

Option A: Claude Code (CLI or VS Code Extension)

Create or edit .mcp.json in your project root (or ~/.claude/mcp.json for global access):

{
  "mcpServers": {
    "datacloud": {
      "type": "stdio",
      "command": "bash",
      "args": [
        "-c",
        "set -a && source /absolute/path/to/data360-mcp/.env && set +a && cd /absolute/path/to/data360-mcp && .venv/bin/python server.py"
      ]
    }
  }
}

Option B: Cursor

Open Cursor Settings > MCP > Add new global MCP server, or edit ~/.cursor/mcp.json:

{
  "mcpServers": {
    "datacloud": {
      "command": "bash",
      "args": [
        "-c",
        "set -a && source /absolute/path/to/data360-mcp/.env && set +a && cd /absolute/path/to/data360-mcp && .venv/bin/python server.py"
      ],
      "disabled": false,
      "autoApprove": [
        "list_tables",
        "describe_table",
        "list_data_streams",
        "list_data_model_objects",
        "list_segments",
        "list_calculated_insights"
      ]
    }
  }
}

Tip: autoApprove lists read-only tools that won't prompt for confirmation each time.

Option C: Claude Desktop

Edit your config file:

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "datacloud": {
      "command": "bash",
      "args": [
        "-c",
        "set -a && source /absolute/path/to/data360-mcp/.env && set +a && cd /absolute/path/to/data360-mcp && .venv/bin/python server.py"
      ]
    }
  }
}

Step 8: Use it

Restart your MCP client (or reload the MCP server list)
Ask it something — for example: "List all tables in Data Cloud"

If the CLI session expires, run sf org login web --alias data360 again. The server does not open its own browser login and does not store a token cache of its own. Delete tools stay refused until the call includes confirm=true after you approve the delete.

Verify It Works

Try these prompts in your MCP client to confirm everything is connected:

Prompt to try	Expected result
"List all tables in Data Cloud"	Returns a list of table names (DMOs, DLOs, CIOs)
"Show me all data streams and their status"	Returns streams with Active/Inactive status
"Describe the ssot__Individual__dlm table"	Returns column names for the Individual DMO
"How many segments do I have?"	Returns a count and list of segment names
"Run this SQL: SELECT COUNT() FROM ssot__Individual__dlm"*	Returns a row count

If any of these fail, check the Troubleshooting section.

Tool Catalog

Query & Schema (3 tools)

Tool	What it does
`query`	Execute SQL (PostgreSQL dialect) against Data Cloud. Results are capped at 1,000 rows and 1 MB unless `MAX_QUERY_ROWS` / `MAX_QUERY_BYTES` are set.
`list_tables`	List all queryable tables. Filterable via `DEFAULT_LIST_TABLE_FILTER` env var.
`describe_table`	Get column names for a specific table.

Data Streams (6 tools)

Tool	What it does
`list_data_streams`	All streams with status, source, category, and last refresh date.
`get_data_stream_info`	Detailed configuration for a single stream.
`refresh_stream`	Trigger a manual data refresh.
`create_new_data_stream`	Create a stream from any connector (CRM, S3, GCS, Azure, SFTP, MuleSoft, IngestApi).
`create_ingestion_schema`	Define the schema for an Ingestion API stream.
`delete_stream`	Remove a stream (and optionally its Data Lake Object).

Data Lake Objects (3 tools)

Tool	What it does
`create_dlo`	Create a DLO with fields, category, and optional dataspace filters.
`update_dlo`	Update a DLO's label or add new fields.
`delete_dlo`	Delete a DLO by record ID or developer name.

Data Model Objects (5 tools)

Tool	What it does
`list_data_model_objects`	All DMOs with metadata.
`get_dmo_details`	Fields and relationships for a specific DMO.
`create_custom_dmo`	Create a fully custom DMO with fields and category.
`create_custom_dmo_fields`	Add custom fields to an existing DMO.
`delete_custom_dmo_fields`	Delete custom fields from a DMO.

DLO-DMO Mappings (5 tools)

Tool	What it does
`list_mappings`	All stream-to-DMO field mappings.
`create_dlo_dmo_mapping`	Create a mapping with pre-flight validation, auto PK/timestamp injection.
`update_dlo_dmo_mapping`	Update an existing mapping (delete + recreate).
`delete_dlo_dmo_mapping`	Delete a mapping by name or by source/target entity.
`batch_create_dlo_dmo_mappings`	Create multiple mappings in parallel (grouped by DLO to avoid lock contention).

Segments (5 tools)

Tool	What it does
`list_segments`	All segments in the org.
`create_segment`	Create a segment from a JSON definition.
`create_segment_dbt`	Create a SQL-based segment (automatically discovers field API names).
`update_segment`	Update a segment (definition, schedule, description, etc.).
`delete_segment`	Delete a segment by API name.

Activations (4 tools)

Tool	What it does
`list_activations`	All segment activations with pagination and ordering.
`get_activation_details`	Full details of a specific activation.
`update_activation`	Update activation settings (refreshType, filters, static data, etc.).
`delete_activation`	Delete an activation by ID.

Connectors (5 tools)

Tool	What it does
`list_connectors`	All SSOT connectors (activation targets, marketing connectors).
`get_connector_metadata`	Metadata for a specific connector type.
`list_available_connectors`	Data source connectors configured in the org.
`list_connector_objects`	Source objects available in a connector.
`list_connected_source_objects`	Objects that already have streams (avoid duplicates).

Data Action Targets (4 tools)

Tool	What it does
`list_data_action_targets`	All data action targets (S3, Marketing Cloud, webhook, etc.).
`get_data_action_target`	Details of a specific target by API name.
`create_data_action_target`	Create a target (Core, MarketingCloud, or WebHook).
`delete_data_action_target`	Delete a target by API name.

Data Actions (2 tools)

Tool	What it does
`list_data_actions`	All data actions configured in Data Cloud.
`create_data_action`	Create a data action triggered by DMO record changes (CREATE/UPDATE/DELETE).

Data Spaces (5 tools)

Tool	What it does
`list_data_spaces`	All data spaces.
`get_data_space`	Details of a specific data space by ID or name.
`update_data_space`	Update a data space's label and/or description.
`list_data_space_members`	All members (DMOs, DLOs, etc.) in a data space.
`get_data_space_member`	Details of a specific member in a data space.

Data Transforms (4 tools)

Tool	What it does
`create_data_transform`	Create a transform (BATCH with STL nodes, or STREAMING with SQL).
`update_data_transform`	Full replacement update of a transform (PUT).
`delete_data_transform`	Delete a transform by name or ID.
`get_data_transform_run_history`	Run history with execution statuses and timestamps.

Calculated Insights (2 tools)

Tool	What it does
`list_calculated_insights`	All calculated insights.
`create_calculated_insight`	Create a CI from a SQL SELECT expression with inferred dimensions/measures.

Data Graphs (4 tools)

Tool	What it does
`list_data_graphs`	All data graph definitions.
`get_data_graphs_metadata`	Full metadata for all data graphs (structure, DMOs, relationships).
`create_data_graph`	Create a graph rooted at a primary DMO with nested relationships.
`delete_data_graph`	Delete a data graph by developer name.

Identity Resolution (1 tool)

Tool	What it does
`list_identity_rulesets`	Identity resolution ruleset configuration.

Retrievers & Search (2 tools)

Tool	What it does
`list_retrievers`	All retrievers (system + custom, for RAG use cases).
`list_search_indexes`	All semantic search index definitions.

Salesforce REST (3 tools)

Tool	What it does
`sf_rest_api`	Read-only GET for `/services/data/.../ssot/...` or an sObject describe. Writes, Tooling API, and SOQL are refused.
`describe_sobject`	Get field metadata for a Salesforce sObject. The name must be a single identifier.
`create_sobject_record`	Create a record on an allowlisted Data Cloud object such as `MarketSegmentDefinition`. Core objects such as Account are refused.

Configuration Reference

Environment Variable	Required	Default	Description
`SF_TARGET_ORG`	No	CLI default org	Alias or username from `sf org login web`.
`DEFAULT_LIST_TABLE_FILTER`	No	`%`	SQL LIKE filter for `list_tables`. Letters, numbers, underscores, and `%` only.
`MAX_QUERY_ROWS`	No	`1000`	Maximum rows returned by `query`.
`MAX_QUERY_BYTES`	No	`1000000`	Maximum serialized response size returned by `query`, including metadata.
`MCP_ENABLE_WRITE_TOOLS`	No	`true`	Set to `false` to refuse every create tool.
`MCP_ENABLE_DESTRUCTIVE_TOOLS`	No	`true`	Set to `false` to refuse every update and delete tool.
`MCP_ALLOWED_WEBHOOK_HOSTS`	For external targets	empty	Comma-separated HTTPS hostnames allowed for webhook and Marketing Cloud endpoints.

Architecture

flowchart LR
    Client["MCP Client<br/>(Claude Code / Cursor / Claude Desktop)"] -->|stdio| Server["server.py<br/>FastMCP"]
    Server --> OAuth["oauth.py<br/>sf org display"]
    Server --> SQL["connect_api_dc_sql.py<br/>SQL query + row cap"]
    Server --> DC["connect_api_datacloud.py<br/>Connect REST API"]
    OAuth -->|access token| SF["Salesforce CLI session"]
    SQL -->|POST /ssot/query-sql| DCAPI["Data Cloud Query API"]
    DC -->|/ssot/*| ConnectAPI["Data Cloud Connect API"]
    DC -->|/sobjects/*| RestAPI["Salesforce REST API"]

Security

Salesforce CLI session — the server uses sf org display for the org named by SF_TARGET_ORG. It does not store a Connected App consumer secret.
Mutations require confirm=true — delete, update, data-action, and data-action-target tools refuse to run until that flag is set after an explicit approval. Read-only tools are marked with readOnlyHint; deletes and updates are marked destructiveHint.
External destinations are allowlisted — webhook and Marketing Cloud endpoints must be https hostnames listed in MCP_ALLOWED_WEBHOOK_HOSTS.
Tool categories can be disabled — set MCP_ENABLE_WRITE_TOOLS=false or MCP_ENABLE_DESTRUCTIVE_TOOLS=false when the connected user should not have that access.
sf_rest_api is GET-only — allowed paths are Data Cloud /ssot/ resources and sObject describe. Tooling API, SOQL, composite, and write methods are refused.
create_sobject_record is allowlisted — only named Data Cloud configuration objects can be created.
Path identifiers are validated — names placed in REST paths must match letters, numbers, underscores, and hyphens, then they are URL-encoded.
Query results are capped — query stops at MAX_QUERY_ROWS. The complete response, including metadata, must fit in MAX_QUERY_BYTES; an oversized row is omitted rather than returned.
Usage logs stay local and private — they are written to ~/.data360-mcp/usage_logs with 0700/0600 permissions. Arguments and errors are stored as SHA-256 hashes.
Rate limits are shared by category — all destructive tools share 10 calls per minute, all writes share 20, and all reads share 120. Override with MCP_DESTRUCTIVE_RATE_LIMIT, MCP_WRITE_RATE_LIMIT, and MCP_READ_RATE_LIMIT.
Tool definitions are pinned — startup compares the tool list with tool_definitions.sha256 and stops on a mismatch.
Local transport only — the server communicates via stdio and does not listen on a network port.

Troubleshooting

Problem	Cause & Fix
`Salesforce CLI not found`	Install the CLI with `npm install -g @salesforce/cli`, then run `sf org login web --alias data360`.
`sf org display` failed	The alias in `SF_TARGET_ORG` is not authenticated. Run `sf org login web --alias data360` again.
`403 Forbidden` from Data Cloud APIs	The logged-in user lacks permission for that Data Cloud operation. Add only the missing permission; do not grant admin access for a read-only setup.
`Invalid identifier`	A name passed to a tool contained `/`, `.`, or another character that is not allowed in a URL path segment. Use the API name only.
Refused to change data	Update, delete, and external-routing tools require `confirm=true` after you explicitly approve the action.
`query` result says truncated	The result hit `MAX_QUERY_ROWS` or `MAX_QUERY_BYTES`. Narrow the SQL or raise those limits.
Tools not showing in Cursor	Click the refresh icon in Cursor MCP settings. Verify `.venv/bin/python -m mcp --version` works.
`401 Unauthorized` mid-session	The CLI access token expired. Run `sf org login web --alias data360` again.
`ModuleNotFoundError: No module named 'mcp'`	Dependencies not installed. Run `pip install -r requirements.txt` in your activated venv.
`Connection refused` in MCP client	The server isn't running or crashed on startup. Check the path in your MCP config and try `python server.py` manually to see error output.

Examples

The examples/ folder has standalone scripts using the same OAuth + API helpers:

# Count active data streams
python examples/count_streams.py

# Dump org DMO metadata
python examples/dump_metadata.py

Useful for verifying the Salesforce CLI login works before wiring up an MCP client.

Contributing

See CONTRIBUTING.md. Bug reports and PRs welcome.

License

Apache License 2.0. See SECURITY.md for vulnerability disclosure.

新增此伺服器

用我們的金鑰與日誌託管一份副本。

Server facts

Tools

Questions