You can connect your AI assistant to Elements in a few minutes. Once connected, your assistant can use the skills described in 'What you can do with Elements in your AI assistant' to answer questions about your Salesforce Org.

Elements runs in more than one location, so the address your assistant connects to depends on which one your Org uses. Match the URL shown in your browser when you're logged into Elements; never guess based on where you or your company are located.

In the steps below, replace <your-api-host> with the host from the right-hand column above, for example api.q9elements.com.

Each AI assistant connects to Elements in a slightly different way. Find your client below and click it to expand the setup steps.

The first time you connect, Elements opens a browser tab and asks you to sign in with your existing Elements account. After signing in, you choose which Space or Spaces the connection can use, then approve. Clients are offered the same full scope set by default, though a client may request fewer; you never pick scopes.

If your account only has access to one Space, Elements selects it automatically. To change which Spaces a connection can use, revoke it and reconnect; see 'MCP security & admin controls'.

Ask your assistant something like "What Elements Spaces can you see?" or "Search my org for a field called…". If you get a real answer back, you're connected.

Then try a first real question:

Some capabilities depend on your Space's plan and licenses; 'Supported tools in the Elements MCP server' lists them, and per-tool requirements are in the tools reference on GitHub. The full menu of capabilities, with example prompts, is in 'What you can do with Elements in your AI assistant'.

Next, read 'How the Elements MCP server works' to understand what it can access, and 'MCP security & admin controls' to see how to manage or revoke a connection. If anything didn't work along the way, see 'MCP troubleshooting' at the end of this article.

The Elements MCP server lets your AI assistant securely query your Elements and Salesforce architecture data, using the Elements permissions you already have. This section explains what happens behind the scenes when you connect.

Your assistant talks to a single Elements address over a secure, authenticated connection. The connection itself stores nothing beyond the sign-in authorization you approved: your password is never shared with your assistant, and no copy of your data is created. There's no separate app or plugin to install on the Elements side; the connection is built into your account.

Diagram: your AI assistant (Claude, Codex, ChatGPT, Cursor, Microsoft Copilot Studio, or any OAuth 2.1 + PKCE MCP client) connects over MCP/HTTPS to the Elements MCP server, which checks your identity, approved Space(s), and role/plan/licenses on every call before reaching your Space. Your Space holds Salesforce metadata synced one-way, read-only, from your Salesforce org. There is no path between your AI assistant and your Salesforce org, ever.

A connection never gives your assistant more access than you already have in Elements. Every request is checked against your Elements role, plan, and Elements permissions at the moment it's made, so if your access changes, what your assistant can see changes with it.

When you connect, you choose which Space or Spaces your assistant can use. It only works in the Spaces you've selected, and a Space must be enabled for the MCP server by Elements before it can be used at all.

Today, your assistant can read your Salesforce metadata and architecture data through Elements: objects, fields, flows, dependencies, permissions, and org-health analytics such as automation, technical debt, compliance, and governance. It can also create and change a limited set of things inside Elements: a new Space, a saved metadata view, a process-map or imported diagram draft, draft user stories, a proposed metadata node (a placeholder marked proposed for people to review, not a change to your org), and MetaFields (your own custom fields on metadata, including their values). Everything else is read-only. Nothing it does ever reaches Salesforce: it never creates, updates, deletes, or deploys Salesforce metadata, and it never overwrites your synced copy. There is no delete tool. The one thing to know is that MetaFields overwrite: populating a classification replaces whatever value was there, and narrowing a MetaField's definition can permanently remove values already held in Elements. Publishing an approved diagram version is deliberately not possible here; it stays a password-confirmed step in the Elements web app.

The Elements MCP server works with any AI assistant whose MCP client supports remote MCP servers with the MCP standard's secure sign-in (OAuth 2.1 with PKCE). It has been tested with Claude (CLI and Cowork), Codex CLI, ChatGPT (Work), and Cursor; see 'Connect your AI assistant to Elements' for setup steps.

This section explains how the Elements MCP server keeps your data safe and how you can govern and audit your connections. It's the section to share with your security team.

Elements stores Salesforce metadata and configuration, along with usage telemetry used by some tools: record counts and Apex execution telemetry from Salesforce Event Monitoring logs. Metadata includes things like object and field definitions, flows, Apex classes, validation rules, and permission sets. It never stores Salesforce business records (accounts, contacts, opportunities, transactions), so nothing your AI assistant does through Elements can reach them either.

The MCP server never connects to your Salesforce org, and nothing it does can change your Salesforce org: Elements reads Salesforce metadata, it never writes it. When your assistant creates or changes something, it writes only Elements artifacts: a Space, a metadata view, MetaFields and their values, a proposed metadata node, a diagram, or a user story.

You sign in with your existing Elements account using OAuth 2.1, an open, secure sign-in standard used across the industry. The sign-in flow uses PKCE, an additional safeguard that prevents an intercepted request from being reused by anyone else. You never share an Elements password with your AI assistant; you sign in through your browser, directly with Elements.

When you approve a connection, you're shown the client (for example, Claude Code) and its requested scope families; clients are offered the same full set by default, though a client may request fewer, and you never pick scopes. The choice you make is which Space or Spaces it can use. Elements controls which Spaces are licensed for MCP access, and your Space administrator controls the Elements permissions you have, which determine what will actually work. A connection only ever reaches the Spaces you selected; new Spaces are never added to an existing connection automatically. To change which Spaces a connection can use, revoke it and reconnect.

Elements turns on MCP access for each Space individually; it is a licensed feature. A Space that hasn't been enabled can't be selected when you connect, and Elements checks this again on every single request, so disabling a Space takes effect immediately, even for connections that were approved earlier.

A connection only ever works inside the Spaces you granted it; it can't be used to reach any other Space, tenant, or customer's data. When no Space is active, Elements uses your Elements default Space if it is among the Spaces you consented to. The active_space_required error appears only when none of your consented Spaces is your default. Elements never uses a Space outside the grant. Behind the scenes, each access token is cryptographically tied to the Elements MCP server, so even if a token were somehow exposed, it couldn't be replayed against any other Elements service.

Access tokens are short-lived and refresh automatically in the background. Refresh tokens rotate on each use and last 30 days from the last refresh, so a connection used at least monthly stays signed in. Elements never stores a token in plain text; only a keyed hash (HMAC-SHA-256) is kept.

You can review and revoke your own connections at any time: open your Elements profile, select the 'Connected apps' tab, find the connection, and click 'Revoke'. Each connection shows the client, the Space(s) it can use, its scope, and when it last signed in. Space administrators can review the connections into their Space and remove their Space from them. Removal takes effect immediately.

Every action your assistant takes through Elements is logged: who made the request, which Space it ran in, what it asked for, and whether it succeeded. Sign-ins, approvals, and revocations are logged too. These records are kept in secure, tamper-resistant storage operated by Elements and used for security monitoring and incident investigation. They aren't currently visible inside the Elements app; if you need a connection's activity investigated, contact Elements support. What you can see yourself, at any time, is every active connection, what it's allowed to do, and when it last signed in, on the 'Connected apps' tab of your profile.

Each connection is limited to a safe rate of requests, and every request is checked against a strict, predefined set of allowed inputs before it runs, so it can't be used to send unexpected commands. Results are screened on the way back, too: hidden characters that could be used to smuggle instructions to an AI assistant are stripped from every response, and AI-generated explanations are checked and clearly marked so your assistant treats them as information to read, not instructions to follow.

Actions that create or change something in Elements still require your own Elements permissions. Your assistant acts as you: it can never do more than your Space administrator has already allowed you to do yourself, and approving a connection never raises your own Elements permissions. And nothing your assistant does ever changes your Salesforce org itself. The most significant changes are saved as drafts for a person to review in Elements: a proposed metadata node is clearly marked as proposed in your reference model, and an imported diagram only updates the diagram's working draft; publishing an approved version is a separate, deliberate step.

Some tools use a large language model to produce answers: explain_metadata_item; the Decision Engine tools run_automation_inventory, run_automation_recommendation, run_schema_inventory, run_schema_recommendation, run_dependency_scan, run_change_impact_analysis, run_architecture_risk_analysis, run_backlog_draft, and get_analysis_result; and generate_process_map and generate_stories_from_diagram. These run through the same licensed AI features Elements already offers in the web application; connecting an assistant doesn't add any new AI integration or send your data anywhere new. Only the metadata needed to answer the question is sent, never Salesforce business records, which Elements doesn't hold. AI-generated narrative is screened and marked as untrusted data. Decision Engine runs keep their analysis as internal Elements artifacts.

The MCP server runs on the same platform as the rest of Elements. Our platform-level security documentation (including SOC 2, ISO 27001:2022, and our Data Processing Agreement) is available at trust.elements.cloud.

Your AI assistant uses a set of underlying tools to answer questions through Elements. This section lists them for reference; you don't need to call them by name, since the skills described in 'What you can do with Elements in your AI assistant' choose the right one automatically.

Every tool respects your existing Elements permissions: a tool never returns more than you could already see in the Elements app. The set of available tools grows over time as Elements adds new skills.

Only a small set of tools write anything (the Read/Write column below shows which). There is no delete tool, and no tool ever mutates Salesforce itself; however, update_metafield can irreversibly remove MetaField values held in Elements when applicability is narrowed or options are removed, and set_metafield_values can clear values, so follow the counting-before-destructive-edit recipe in the elements-metafields skill before making those changes.

Some tools are only available on certain Elements plans, such as Pro or Enterprise, or to users with a required role. A tool that no authorized Space makes available through its plan, licenses, and your role won't appear in your assistant's tool list at all; if you call a visible tool in a Space that doesn't qualify, or without the required rights to the specific resource, your assistant lets you know.

These tools exist for the Remove Unused Apex playbook and are described on its page. Apex execution telemetry comes from Salesforce Event Monitoring daily Apex event logs; no telemetry does not mean nothing is used.

Data load impact analysis uses fully synced configuration to show which automations, validations, and locking consequences a planned Salesforce data load can trigger before it runs.

Analysis runs in the background: poll get_analysis_result before starting the next step. Elements stores analysis artifacts; requirements and stories remain drafts. Schema creation goes directly from an accepted create-new option to backlog; reuse and extension include dependency, impact, and risk analysis.

Elements connects to your AI assistant so you can ask questions about your Salesforce Org in plain language, right inside the tools you already use. Built-in skills turn a plain-language request into the right Elements action, so you get an answer instead of digging through screens.

These fifteen skills are available today. Each one is built to recognize the kind of question you're asking and call the right Elements lookup automatically; you don't need to learn any special commands. If your client has the Elements skills installed (see 'Connect your AI assistant to Elements' above), you can also invoke a skill by name: use the slash form shown with each example, or $ in Codex (for example $elements-spaces).

Ask your assistant which Elements Space it's currently using, switch it to a different Space, or create a new one for a project.

"What Space are you working in right now? Switch to the Acme Corp space."
or directly: /elements-spaces switch to Acme Corp

Ask about a specific object, field, flow, trigger, or Apex class: what it is, what it depends on, and what it does.

"What does the LoyaltyPointsFlow do, and what does it depend on?"
or directly: /elements-metadata LoyaltyPointsFlow

Ask which Profiles, Permission Sets, Permission Set Groups, or Groups can read or edit a field or object, and how many users each one reaches.

"Who can edit the Credit Limit field on Account, and how many people does each permission reach?"
or directly: /elements-access Credit Limit on Account

Ask for a filtered list of metadata that matches specific criteria, and save it as a named, reusable view.

"List every screen flow still on an old API version, and save that as a view called 'Flows to upgrade.'"
or directly: /elements-metadata-query screen flows on old API versions

Ask for a full health check of your org (structure, automation, technical debt, compliance, and governance) in one pass, with a ranked summary of where to look first.

"Run a health check on my org and tell me where to look first."
or directly: /elements-org-diagnostic

Ask what changed in your org over a given period, broken down by metadata type, with unusual spikes and automation bursts flagged.

"What changed in my org this week, and was anything unusual?"
or directly: /elements-change-briefing last 7 days

Ask for a ranked list of what to fix or remove first, based on risk, how many other components depend on it, and how recently it changed.

"What should I clean up first, and why?"
or directly: /elements-tech-debt

Ask your assistant to scan your whole Org for agentic opportunities and rank them, with the evidence behind each, so you know where an Agentforce agent would pay off.

"Scan my Org and tell me the top few places we should use an Agentforce agent."
or directly: /elements-agent-opportunities

Ask what it would take to uninstall a managed package. Elements inventories the package's components (as completely as your Space's licensed tools allow), sweeps them for incoming dependencies, separates the real blockers from references that disappear on their own, and hands back a sequenced removal plan, from resolving blockers through a sandbox dry run to post-removal verification. It plans the removal; the uninstall itself stays in your hands.

"What would it take to remove the Consensus package from our org?"
or directly: /elements-package-uninstall Consensus

Every skill above answers about one org at a time. This one composes those answers into a comparison across all the orgs in your Space.

Ask for a view across every org you can see. Elements discovers the orgs in your Space, asks you to confirm the list, runs the same query against each one, and produces a self-contained HTML report you can email to someone who has no Elements access and no AI assistant. Every number on the page traces back to the raw response that produced it, and each org carries the timestamp of its last successful sync, so an org that has quietly stopped syncing is flagged rather than averaged in. A worked inventory report ships with it, and you can customize it by describing what you actually want; a technical-debt comparison is available as a data pull while its report is being redesigned. Because each run saves a dated snapshot, running it again next week turns the same report into a trend.

This one needs an assistant that can run Python: the report is produced by a script rather than written by the model, which is what makes the numbers reproducible.

"Give me an overview of our Salesforce orgs."
or directly: /elements-multi-org-analytics

Skills for structuring and classifying metadata, and evaluating how to make a change before you build it.

Ask your assistant to set up the custom fields your Space uses to classify Salesforce metadata, connect dependent picklists, populate values in bulk, and verify the result. Definition changes stay in Elements, but narrowing applicability or removing options can purge existing MetaField values across synced Orgs.

"Set up a retention-priority MetaField for our objects, classify them, and verify the result."
or directly: /elements-metafields classify our objects by retention priority

Ask for the process behind a lifecycle object such as Opportunity, Case, or Lead. Your assistant builds it from the org's actual record types, lifecycle fields, and synced automations rather than a generic template, producing one diagram per record type.

"Show me the Case process."
or directly: /elements-org-to-process Case

Ask how best to add or change Salesforce automation, objects, fields, or record types. Your assistant inventories existing metadata and recommends approaches. Once you select one, it analyzes existing targets and drafts a delivery backlog. A create-new schema selection goes directly to the draft. You receive a chat summary and local report; Elements stores analysis artifacts while requirements and stories remain drafts.

"We want Cases auto-assigned by region. What's the best way to build it and what would it affect?"
or directly: /elements-decision-engine

Skills for turning a decision into diagrams and artifacts.

Ask your assistant to turn a description, transcript, or images into a UPN process map, or to read and edit an existing diagram. Edited diagrams are saved as a draft for you to review and release in Elements.

"Turn this onboarding call transcript into a process map, then add an approval step before go-live."
or directly: /elements-diagrams

Ask your assistant to list the user stories in a Space, or to generate stories from the activities on a process diagram, each with a headline, description, and acceptance criteria.

"Generate user stories from the activities on this onboarding diagram."
or directly: /elements-stories

Elements keeps extending the MCP server. New skills appear here as they ship.

Playbooks are separate plugins that run a complete Salesforce process end to end. Read 'What is an MCP playbook?' to learn how they work and how to install one.

If your AI assistant can't connect to Elements, or something isn't behaving as expected, start here. Click any question below to expand it.