AI integration · MCP server

Run Manifestly from any AI assistant.

Our MCP server connects your workflows to Claude, ChatGPT, Gemini, Cursor, and every MCP-compatible client — so "what's overdue?" and "start onboarding for Acme" become things you just ask.

Endpoint
mcp.manifest.ly
Transport
Streamable HTTP
Auth
OAuth 2.1
Setup
~ 60 seconds
Background

What is MCP?

MCP — the Model Context Protocol — is the open standard, originally built by Anthropic, that lets AI assistants securely connect to outside tools.

Think of it as a universal port. Build it once, and every MCP-compatible AI can use it. We've built it for Manifestly. Plug in the assistant of your choice and it can read, run, and reshape your workflows.

How Manifestly connects to AI assistants via MCP
What you can do

Three things you'll do every day.

We've watched dozens of teams use this. The same three patterns keep coming back: a morning habit, an in-the-flow habit, and a once-in-a-while wow.

01 / Daily habit

Stay on top of the work without opening Manifestly.

Start your morning by asking. Get a real-time read on overdue steps, late runs, and what your team is actually doing, pulled live from your account, scoped to what you have permission to see.

For: managers · ops leads · standups
Claude · Manifestly connected
You
What's overdue across the ops team this week?
Claude
Pulling overdue assignments for the ops team.
→ list_my_assignments(team=ops, status=overdue)
4 overdue across 3 runs:
• Acme onboarding · "Provision SSO" · Priya · 2d late
• Acme onboarding · "Send welcome packet" · Jordan · 1d late
• Q2 SOC2 evidence · "Collect access reviews" · Sam · 5d late
• Northwind kickoff · "Schedule kickoff call" · Priya · 1d late
You
Reassign Priya's two to Marcus and DM her a heads-up.
Claude
Done. Reassigned both steps to Marcus, drafted a Slack DM to Priya — want me to send it?
02 / In the flow

Run the workflow without leaving the conversation.

Kick off runs from templates, complete steps, assign work, set due dates, fill in data fields, leave comments. Every action your team takes in Manifestly is now a sentence away. No tab-switching, no context loss.

For: anyone executing real work
ChatGPT · Manifestly connected
You
Start the new client onboarding workflow for Northwind. Mark step 1 complete and assign the contract review to Priya.
ChatGPT
Starting the run, completing step 1, and assigning the contract review.
→ start_run_by_workflow_name("Client onboarding")
→ complete_run_step(run=14228, step=1)
→ assign_run_step_by_name(run=14228, step="Contract review", user="Priya")
✓ Run #14228 — "Northwind onboarding" started
✓ Step 1 "Create CRM record" — completed
✓ Step 4 "Contract review" — assigned to Priya Mehta, due Fri
You
Add a comment on the security review step: "Northwind needs SOC2 type 2, see their RFP, page 14."
ChatGPT
Comment added on step 7. Anything else?
03 / The wow

Design new workflows by describing them.

Most workflow tools make you build templates by hand. Ours gives the AI the schema. Describe what you need, including phases, steps, owners, cadence, and conditional logic, and it generates a complete, valid template, validates it, and imports it into your account.

For: turning SOPs into runnable workflows in minutes
Claude · Manifestly connected
You
Build me a SOC2 evidence collection workflow with monthly cadence: collect access reviews, run vulnerability scans, document changes, and route everything to the security lead.
Claude
Drafting the template against your account's schema, then validating before import.
→ get_workflow_schema()
→ validate_workflow_import(template=...)
→ import_workflow(department="Security", template=...)
✓ Workflow created: "SOC2 — Monthly Evidence"
   Steps: 11 across 4 phases
   Owners: Security Lead (default), reassignable per run
   Schedule: Monthly, day 1
   Imported into department: Security
You
Looks good. Schedule the first run for next Monday.
Works with

Every modern AI client. One server.

We host the integration. They speak MCP. You connect once and use whichever assistant your team prefers.

Claude
Claude
Web · Desktop · Code
Coming soon
ChatGPT
ChatGPT
Pro · Business · Enterprise
Gemini
Gemini
Enterprise · Agent Designer
Cursor
IDE
Anywhere MCP runs
Custom agents · n8n · more

If your AI tool supports MCP, it works with Manifestly.
New clients add MCP support all the time, you don't have to wait for us to build a new integration.

Capability surface

Not a read-only connector. A full surface.

Most B2B MCP servers let AI ask questions about your data. Ours lets AI do the work: read, write, and even author new workflows. Everything respects your existing roles and permissions.

Read

Answer almost any "what's the status of..." question.

  • → Workflows + run steps
  • → Runs (active, completed)
  • → Users + departments
  • → Comments + data fields
  • → My assignments
  • → Webhook registry

Write

Take the action your team would take in the UI.

  • → Start runs from templates
  • → Complete · skip · uncomplete
  • → Assign by name or email
  • → Set due dates + fields
  • → Add comments
  • → Manage webhooks

Author

Generate, validate, and import full workflow templates from prose.

  • → Get workflow schema
  • → Validate import
  • → Import workflow
  • → Export workflow as JSON
  • → Round-trip + version
  • → Schedule + tag

Every action is attributed to the human who asked. Permissions are always honored.

Setup

Three ways to connect. Roughly a minute.

We host the server, so there's nothing to install. Pick your client, paste the URL, sign in.

Step 01 · Most clients

Claude, ChatGPT, Cursor

Add Manifestly as a custom MCP connector. Sign in with your Manifestly account when prompted. Done.

https://mcp.manifest.ly
Quick-start guide »
Step 02 · Enterprise

Gemini Enterprise

Your Google Workspace admin wires Manifestly into Agent Designer as a custom MCP data source. We have the IAM and OAuth steps documented.

Admin setup required
Admin setup guide »
Step 03 · Custom

Build your own agent

Wire Manifestly into your own MCP-compatible client, n8n flow, or custom agent. Streamable HTTP, OAuth 2.1, standard plumbing.

See API reference
Developer docs »
Built right

Boring where it matters.

AI integrations should not be a security afterthought. We built this to enterprise standards from day one.

— Auth
OAuth 2.1, no API keys to share.

Users sign in with their Manifestly account. We never see your AI provider's credentials, and you never paste a long-lived secret.

— Permissions
Scoped to the user who connected.

The AI can only see what that user can see, and only do what that user can do. Your existing roles and departments apply.

— Transport
Modern Streamable HTTP.

The current MCP standard, not the deprecated SSE pattern. Works with every recent client and every enterprise admin tool.

— Audit
Every action attributed.

When the AI completes a step or starts a run, the audit log records the human who asked. No "AI did it" mystery actions in your history.

Manage connected apps in your Manifestly profile »

Talk to your workflows.

Your runbooks are about to become things you run by asking. Connect Manifestly to your AI client in under a minute.