Tell the assistant what you need to track and what should happen to it, and it builds the record types and the workflow for you.
Use the built-in AI Assistant, or bring the same workspace tools into Claude, ChatGPT, Codex and Cursor through MCP.
It works inside your account as you, not as a service account — so it can only reach what your own permissions already allow, and every action it takes lands in the same audit trail as if you had clicked it yourself.
Use your own AI client through MCPThe shortest way to understand it is to see the exchange.
Every action above is one you could have taken yourself, and each lands in the audit trail under your name.
Nothing goes live because the assistant decided it should. It drafts, it explains what it is about to do, and the publishing and activating stay with you — which is also why letting it try something is cheap.
The assistant drives the products' own operations, so its limits are your permissions rather than a shorter feature list.
Not a separate assistant-only feature list — these are the products' own operations, so it reaches what you reach.
Your AI client. Your workspace.
MCP (Model Context Protocol) connects Claude, ChatGPT, Codex, Cursor and other compatible AI clients to your SynaptaGrid workspace. Ask a question or describe a task, and the client uses the tools your account can access.
Inspect record types, fields and schemas. Find records, then make the changes you request.
Inspect workflows and runs, build steps and mappings, and investigate a failed run.
Find available external systems and read their activity contracts before configuring a workflow.
Read activity history to understand what happened and who performed an action.
Read support conversations and record discussions in the workspace you connected.
Check your inbox. Template management is available when your workspace enables it.
Use SynaptaGrid’s built-in AI chat or connect your own client. Each tool call runs with your signed-in account. Read the audit history to review recorded workspace activity. Available tools depend on your enabled products, features and permissions. Follow any approval prompts before an action continues.
Start with a specific request: “Show my record types”, “Check the latest workflow runs”, or “Read this support ticket”. For changes, name the target and the result you want. Confirm the connected workspace before working with its data.
Read the MCP connection guideGet the skill and client instructionsThe assistant is in early access. The capabilities below are the ones in customers’ hands today.
It builds
Record types, workflows, connections to systems you already run, and tools of your own — created from a conversation and published for you to review.
Say what you need to keep track of, and the record types, field groups and typed fields are created and published for you.
Describe the process and it creates the steps, the branching, the connections between them, and a version you can publish.
Give it a service’s API description and it registers the system, imports the operations as usable steps, and writes the field mappings.
Define a single-purpose tool with typed inputs and outputs, and it becomes both something you can run and a named step available to your workflows.
It reads
It reads your structure, your versions and your audit trail together, and it reads the documents you hand it — so a question about your account gets an answer grounded in your account.
A read-only assistant correlates the workflow’s structure, its validation state, its published versions and the audit trail to explain what happened.
Upload a file and it reads it, so extracting structure from a real document does not start with retyping it.
It stays inside the lines
It acts as the person using it, never as a service account, and every run is bounded by a budget you set. This is the part a security review asks about.
Every action is checked four times before it runs: your organization’s grant, your personal grant, the assistant’s own allow-list, and a live permission check on the service being asked.
The platform is a remote MCP server, so an AI client can read your records and run your workflows as you, under your own sign-in.
MCP — the Model Context Protocol — is how AI clients connect to external systems. Pointing one at your account is a sign-in, not a key: the client discovers where to authenticate and registers itself, you approve it as yourself, and from then on it sees exactly the tools your own permissions allow. Nothing is exposed by connecting that you could not already do.
Usage draws from a credit pool you top up, with per-person caps, warnings as you approach the limit, and a ceiling on any single run.
Teach your client the workflow
The connection provides live tools. The skill explains which product to use, how to select the workspace, and how to check results and handle access limits. Connect your account in Portal → AI → Connect MCP first; installing the skill does not grant access.
Skills require code execution. If upload is unavailable, check your Claude organization’s skill policy.
Claude skill instructionsegav-mcp folder into ~/.claude/skills/. The file should be ~/.claude/skills/egav-mcp/SKILL.md./egav-mcp with your request.After connecting MCP, extract the ZIP’s egav-mcp folder into ~/.agents/skills/. Both clients can discover ~/.agents/skills/egav-mcp/SKILL.md.
In Codex CLI or the IDE extension, use $egav-mcp. In Cursor, select /egav-mcp. Restart the client if the skill does not appear.
If your client supports file attachments but cannot install a standalone skill, download the instruction file and attach it as chat context. Ask the client to follow it for the current task. This provides guidance for that conversation.
Use Portal’s server URL and sign-in flow to connect. The instruction file cannot connect an account or enable a disabled feature.
Use the SynaptaGrid MCP skill to list the record types and workflows in my connected workspace. Confirm the workspace and report what you actually find. Keep this check read-only.
The quickest way to judge it is to describe a form you already use and watch what it produces.