Guide / By Rob Lowry, Founder of LaunchEngine

Published 2026-07-08 / Updated 2026-07-08

AI Agents on monday.com: The Complete Guide

The definitive reference. Native monday agents (Risk Analyzer, Lead Scorer, Meeting Summarizer, Ticket Assignment, Sentiment Detector, monday sidekick), the AI Agent Builder, and custom development with MCP + GraphQL + monday Code. Data model, integration surfaces, agent-specific rate limits and credits, trigger patterns, and when to use each layer.

See LaunchEngine Agents Or jump to the glossary

01 / Summary

Key takeaways

01AI agents on monday.com are software that works inside the monday data model (Workspaces > Boards > Groups > Items > Subitems). Agents can be assigned to items and mentioned in updates like a team member.
02Three categories exist today: native monday agents (Risk Analyzer, Lead Scorer, Meeting Summarizer, Ticket Assignment, Sentiment Detector, plus monday sidekick), agents built with the monday AI Agent Builder, and custom agents built by developers using MCP servers and the GraphQL API.
03Custom agents get their own account infrastructure via a dedicated Agent Signup API and higher rate limits: 20M complexity points per minute vs 5M for standard integrations.
04Two MCP servers cover the developer surface. Platform MCP (hosted at mcp.monday.com/mcp) reads and writes monday data. Apps MCP (local) manages the app and agent lifecycle: scaffolding, feature creation, deploy.
05Every agent on monday needs six things designed: trigger, tool set, prompt, model, cost telemetry, and gates. The choice between native, Builder, or custom depends on which of the six need to be under your control.
06Native monday agents work out of the box for the standard patterns they cover. The AI Agent Builder handles custom logic without code. Custom-built agents are the right fit when the workflow needs a specific model, a specific prompt structure, per-run cost telemetry, or logic none of the above covers.
07AI credits meter agent-driven work. Every run consumes credits from the account's usage dashboard. Cost visibility matters as agent count scales.
08The marketplace reaches 245,000+ monday.com customers. Custom agents can ship there through the standard app publishing flow, with monday's native monetization and payment processing.

02 / Definition

What an AI agent on monday.com is

An AI agent on monday.com is software that acts inside the monday data model. It triggers on a specific event, reads context from workspaces and boards, calls a large language model to make a judgment or generate output, and writes the results back to items. Custom agents that register through the Custom Agents surface also get first-class platform treatment: they can be assigned to items and mentioned in updates using @, exactly like a team member.

monday brands the category "monday agents" and describes it as an unlimited workforce of ready-made or custom agents that act where the team already works. That framing captures the point: the agent runs inside the workspace the team uses, not in a separate app.

Three layers exist today. Native monday agents. The AI Agent Builder. Custom development. Each has a different tradeoff between speed to first agent, ceiling on customization, and control over prompt, model, and cost.

This guide walks the landscape, the data model and integration surfaces, the developer stack for custom builds, the agent-specific platform capabilities, trigger patterns, the decision framework for choosing a layer, the anatomy of a well-designed custom agent, where to start, common pitfalls, cost model, and a glossary.

03 / Landscape

Three categories of AI agents on monday

The three exist for different needs. Time to first agent is fastest with native; ceiling on customization is highest with custom builds.

Pre-built by monday

Native monday agents

monday ships a growing library of pre-built agents: Risk Analyzer (detects schedule, dependency, and workload risks across projects), Lead Scorer (scores leads using fit, intent, and engagement signals), Meeting Summarizer (creates meeting notes, transcripts, and summaries), Ticket Assignment (detects ticket intent, urgency, and required expertise), and Sentiment Detector (detects sentiment shifts across tickets, emails, and feedback). Plus monday sidekick, the general AI assistant embedded across the platform.

Build without code

AI Agent Builder

monday's own low-code interface for building custom agents. Three-step flow: describe what you need, connect knowledge and tools, deploy. Sits under the broader Agent Factory feature that some monday customers call the Digital Workforce. Right for teams that need behavior beyond what native agents cover but do not have a developer building against the platform APIs.

Developer-shipped, full control

Custom-built agents

Where LaunchEngine builds

Agents built by developers using the monday MCP servers, the GraphQL API, and typed SDKs. Hosted either on monday Code (monday's managed hosting) or an external runtime (Render, AWS, GCP). Full control over prompt, model, trigger logic, tool access, and cost telemetry. Right for specialized workflows, per-client customization at scale, or when a specific model or telemetry story is required.

04 / Native library

The pre-built agents monday ships

Named, out-of-the-box agents. Fastest path to value when a workflow maps to one of these.

Native agent

Risk Analyzer

Detects schedule, dependency, and workload risks across projects in real time.

Use case: Project managers running multiple concurrent workstreams.

Native agent

Lead Scorer

Scores leads using fit, intent, and engagement signals across your funnel.

Use case: Sales teams running CRM boards.

Native agent

Meeting Summarizer

Creates meeting notes, transcripts, and summaries.

Use case: Any team using monday to track meetings and follow-ups.

Native agent

Ticket Assignment

Detects ticket intent, urgency, and required expertise. Routes to the right owner.

Use case: Support and IT teams running ticket boards.

Native agent

Sentiment Detector

Detects sentiment shifts across tickets, emails, and feedback.

Use case: Customer success and CX teams tracking health signals.

Native agent

monday sidekick

General AI assistant embedded across the platform. Answers questions, summarizes, drafts.

Use case: Any monday user who wants inline AI assistance while working.

05 / Data model

How agents work inside monday's structure

Workspaces > Boards > Groups > Items > Subitems. Plus updates. Agents read and write across these primitives.

Primitive	How agents use it
Workspaces	Top level container. Agents can be scoped to specific workspaces per admin configuration.
Boards	Where structured work lives. Agents read board columns, updates, and files, and write results back to items.
Groups	Sections within a board. Agents can trigger on items being created in a specific group (e.g., 'New' or 'To Triage').
Items	The core unit of work. Agents can be assigned to items like a person, run against a single item, or process items in bulk.
Subitems	Nested items. Agents that manage complex workflows often decompose work by creating subitems tied to a parent.
Updates	Threaded comments on items. Agents can be mentioned in updates and respond, and can post their own updates as an audit trail of what they did.

06 / Surfaces

The six integration surfaces

Where agents show up in monday's UI. Which one you build against depends on how the user interacts with the agent.

Board Views

Custom UI embedded on a board. Right surface when the agent has a dashboard, summary view, or configuration interface the user needs to see.

Widgets

Embed on monday dashboards. Right surface for cross-board summaries, KPIs the agent computes, or portfolio-level visualizations.

Menu Items

Add entries to app menus. Right surface for one-shot actions the user invokes from a menu (run a scan, generate a report, launch a workflow).

Doc Actions

Interact with monday docs. Right surface for agents that read, generate, or transform document content.

Custom Agents

Register as an agent surface. Formally recognized by the platform as an agent, gets the account infra (Signup API, higher rate limits, assignment).

Automations

Trigger from monday's automation recipes. Right surface for event-driven work: item created, status changed, column value updated, deadline reached.

07 / Developer stack

What you use to build a custom agent

Seven building blocks. Most builds pull from four or five of them.

Platform MCP

Hosted MCP server at https://mcp.monday.com/mcp. Read and write monday data. Transport: Streamable HTTP. Authentication: OAuth 2.1 or Bearer token with a personal API token. Covers querying boards, creating items, updating columns, managing users.

Apps MCP

Local MCP server launched via `npx @mondaydotcomorg/monday-api-mcp -t <token> --mode apps`. Manages the app and agent lifecycle: app scaffolding, feature creation, deploy, plus semantic search across the developer docs.

GraphQL API

The primary data layer. Direct access for advanced use cases where the SDK does not cover a specific query. Full read/write across the data model. Complexity points meter the rate limit.

SDKs (JavaScript + Python)

Typed SDKs that wrap the GraphQL API. Handle rate limits automatically with configurable retry logic. Best default for most agent builds.

monday Code

monday's managed hosting for apps and agents. Deploy with `mapps code:push`. Includes managed storage, secrets, and observability. Alternative to hosting on Render, AWS, or GCP.

monday CLI (mapps)

Command-line interface for scaffolding, developing, and deploying apps and agents. Pairs with monday Code.

Vibe design system

monday's design system. Ensures UI surfaces feel native to the platform. Use when building Board Views, Widgets, or any surface a user sees.

08 / Agent-specific

What Custom Agents get that standard integrations don't

Registering as an agent unlocks account infrastructure. Higher rate limits, item assignment, update mentions, and dedicated onboarding.

Capability

Agent Signup API

Programmatic account creation for agents at `/signup-logic/agent-signup/account/v1`. Purpose-built onboarding so agents get their own identity in the platform rather than borrowing a user's session.

Capability

Dedicated rate limits

Custom Agents receive 20M complexity points per minute vs 5M for standard integrations. Designed to accommodate high-throughput agent workloads.

Capability

Item assignment

Agents can be assigned to items exactly like a person is. The agent becomes the accountable owner, notifications flow through, and the item's history reflects agent-taken actions.

Capability

Update mentions

Team members can mention an agent in an update using `@`. The agent reads the mention as an instruction and can respond or act. Enables inline delegation without leaving the board.

Capability

AI credits

Agent-driven work consumes credits from the account's usage dashboard. Visible per account. Custom-built agents on external runtimes track their own model API costs (Anthropic, OpenAI, etc.); native agents and Builder agents use monday's AI credits.

Capability

Scoped access

Agents can be granted access to specific workspaces, boards, or item scopes. Admin configuration controls what the agent can see and change. Same permission model that governs human users.

09 / Triggers

How custom agents fire

Five patterns. Most custom agents pick one; multi-pattern agents are usually two agents.

Webhook from a monday automation

The most common pattern for custom agents. A monday automation fires on an event (item created, status changed, column value updated, deadline reached), posts to your agent's webhook, and the agent reads the payload, does the work, and writes back through the GraphQL API.

Managed provider integration

monday supports managed integrations with providers like the Claude Managed Agent. Reduces the auth and lifecycle work for teams standing up their first custom agent.

Scheduled (cron)

Time-based triggers for digest-style agents, periodic scans, or scheduled reports. The agent runs on its own schedule, reads context from monday, and writes results back.

Update mention

A team member mentions the agent in an update. The agent reads the mention as an instruction, decides whether it can act, and responds in the same thread. Enables inline delegation.

External event with signature verification

Incoming email, form submission, or third-party webhook triggers the agent. The agent parses the event, matches it to a monday item or creates a new one, and writes structured data back.

10 / Decision

Native vs Builder vs Custom

Pick the layer that matches the workflow. Do not default to custom when native or Builder covers it.

Dimension	Native	Builder	Custom
Time to first agent	Minutes	Hours	Days to weeks
Customization ceiling	Fixed by monday	Broad but bounded by Builder features	Full control
Prompt control	Vendor-owned	Guided by Builder	Yours, versioned per client
Model control	Vendor-selected	Vendor-selected	Yours (Sonnet 4.6, Haiku 4.5, GPT, or whatever your runtime supports)
Cost visibility	AI credits pool	AI credits pool	Per-run tokens and dollars in your dashboard
Data access	Native monday data	Native + connected knowledge/tools	Anything the MCP or API can reach
Trigger flexibility	Standard triggers	Automation-connected triggers	Webhooks, cron, external events, managed provider
Best for	Common patterns already covered	Custom logic without code	Specialized workflows, per-client customization, specific model or telemetry needs

Check the native library

If a native monday agent already covers your workflow (project risk, lead scoring, meeting summary, ticket routing, sentiment tracking, general AI assistance), start there. Fastest path to value, no code, no runtime to maintain.

Try the AI Agent Builder

For workflows the native agents do not cover but that stay within monday's Builder envelope (connect knowledge, connect tools, describe rules), the Agent Builder is the next step. No developer required.

Build custom when the workflow needs it

Build a custom agent when you need a specific model, a specific prompt structure, per-run cost telemetry, external system integration beyond monday's connectors, or per-client customization at scale. Use monday Code for managed hosting, or an external runtime if you already have one.

11 / Anatomy

What a good custom agent needs designed

Six elements. Every element applies to native and Builder agents too, but on those tiers most elements are vendor-selected.

Trigger

How the agent fires. Webhook from a monday automation, managed provider integration, scheduled cron, update mention, or external event. Every agent has exactly one trigger; multi-trigger agents are usually two agents.

Tool set

What the agent can read and write. For monday-native work, that is the Platform MCP or the GraphQL API. For cross-system work, additional tools (Gmail, external APIs, vision) get added explicitly. Locked at design time to reduce blast radius.

Prompt

The system prompt the agent runs with. Encodes your specific procedure, your voice, and the constraints the agent operates under. Versioned per agent. Ideally editable from a dashboard without a code deploy.

Model

The LLM the agent calls. Reasoning work goes to a strong model (Sonnet 4.6, GPT-5, Gemini Pro). Cheap classification and routing can go to a smaller model (Haiku 4.5). Tunable per agent.

Cost telemetry

Every run logs input tokens, output tokens, dollar cost, latency, model, and status. Native and Builder agents use monday's AI credits pool; custom agents on external runtimes track their own model API costs in their dashboard.

Gates

Which action classes require human confirmation before firing. Outbound comms to customers, large approvals, and any cross-system write to a system of record should start gated. Confirmation rates guide when each class ungates.

12 / Where to start

Map the workflow, pick the layer, ship in observer mode

Same three-step pattern for every layer. Skipping any step is where most agent programs stall.

Map the workflow

Write down the workflow the agent will run. Trigger event, inputs, decisions, outbound actions, gates, and confirmations. This becomes the specification.

Pick the layer

Given the workflow, decide: native, Builder, or custom. Use the decision framework above. Do not default to custom when native or Builder covers it.

Ship in observer mode

On the first pass, every outbound action is gated for human confirmation. Track confirmation rates. Ungate one action class at a time as rates prove out.

13 / Costs

What each layer costs to run

Two cost primitives: AI credits (for native and Builder), and model API + hosting dollars (for custom on external runtimes). monday Code sits between.

Cost model

Native monday agents

Consume AI credits from the account's usage dashboard. Metered per run. Pricing on the account's plan tier. Predictable if usage stays inside the plan; overages meter separately.

Cost model

AI Agent Builder agents

Use the same AI credits pool as native agents. Custom logic built in the Builder consumes credits per run just like a pre-built.

Cost model

Custom-built agents (external runtime)

Model API costs go directly to the provider (Anthropic, OpenAI, etc.). Hosting costs go to the runtime provider (Render, AWS, GCP). Per-run visibility depends on the runtime's telemetry. LaunchEngine Agents logs input tokens, output tokens, dollars, and latency per run.

Cost model

Custom-built agents (monday Code)

monday Code is monday's managed hosting for apps and agents. Consumes AI credits for platform-native AI calls. External model calls made from a monday Code agent hit your provider account directly.

Cost model

Marketplace

Publish through the standard app submission flow. Reach 245,000+ monday.com customers. Native monetization handles payment processing. Marketplace revenue share applies to paid apps.

14 / Pitfalls

Common mistakes

Most of these come from defaulting to the wrong layer, or from skipping observer mode when a new agent goes live.

Building custom when a native agent covers the workflow

The native library keeps growing. Check first. Building a custom agent for something Risk Analyzer or Lead Scorer already does is money and time spent on parity work.

Using the Builder when the workflow needs a specific model

The Builder is fast for custom logic but the model choice is not yours. If you need a specific model (Sonnet 4.6, Haiku 4.5) or a specific reasoning depth, a custom build is the right layer.

Skipping observer mode on a new custom agent

Letting a new agent send comms or write to a system of record on day one is how trust breaks. Start gated, ungate one action class at a time as confirmation rates prove it out.

Ignoring rate limits

Custom Agents get 20M complexity points per minute, but a poorly-written agent can burn through that quickly on high-throughput workloads. Design queries to be complexity-aware and use the SDK's built-in retry logic.

Not scoping AI credit or model API usage

Every run costs. Native and Builder agents consume the account's AI credits; custom agents on external runtimes consume model API dollars. Instrument cost per run per agent from day one.

No handback path for uncertain runs

The agent will encounter cases it cannot handle. Without a structured way to post the item back to a human inbox with context, the workflow stalls silently.

Building outside the workspace the team uses

monday agents work because they run where the team already works. If your custom agent lives in a separate app, it adds friction. Write results back to monday even when the compute happens elsewhere.

No audit trail for agent-taken actions

Every automated write should leave a reviewable record on the monday item update thread. Silent writes make agent behavior impossible to defend later.

15 / Reference

Glossary

The vocabulary of AI agents on monday.com, defined in one place.

AI agent (on monday.com)

Software that acts inside monday's data model. Triggers on an event, reads context from workspaces and boards, calls an LLM to make a judgment or generate output, writes results back to items, and can be assigned to items or mentioned in updates like a team member.

monday agents

monday's brand name for the AI agents category on their platform. Positioned as an 'unlimited workforce' of ready-made or custom agents that act where the team already works.

Custom Agents

One of the integration surfaces on the monday Apps Framework. Registers an app as an agent, which unlocks the account infrastructure (Signup API, higher rate limits, item assignment, update mentions).

AI Agent Builder

monday's low-code builder for creating custom agents. Three-step flow: describe needs, connect knowledge and tools, deploy.

Agent Factory / Digital Workforce

monday's broader framing for orchestrating multiple agents together. Some customers call the collection of agents they build the Digital Workforce.

monday sidekick

The general AI assistant embedded across the platform. Not a specialized agent; a horizontal assistant available in most product surfaces.

MCP (Model Context Protocol)

An open protocol for connecting AI agents to tools and data sources. monday provides two MCP servers.

Platform MCP

monday's hosted MCP server at https://mcp.monday.com/mcp. Reads and writes monday data. OAuth 2.1 or Bearer token authentication.

Apps MCP

monday's local MCP server for the app and agent lifecycle. Launched via `npx @mondaydotcomorg/monday-api-mcp --mode apps`. Handles scaffolding, feature creation, and deploy.

monday Code

monday's managed hosting for apps and agents. Deploy with the mapps CLI. Alternative to running your agent on Render, AWS, or GCP.

Vibe (design system)

monday's design system. Ensures apps and agents that render UI surfaces feel native to the platform.

AI credits

The metering unit for native and Builder agents. Consumed per run from the account's usage dashboard. Included on plan tiers with overage pricing.

Complexity points

The GraphQL rate-limit unit. Standard integrations get 5M complexity points per minute; custom Agents get 20M.

16 / Keep going

Related guides

This is the monday.com AI Agents top hub. From here, dive into vertical applications and implementation deep-dives.

AI Agents for Property Management: The Complete Guide

Vertical-specific application. Six agent roles that cover a PM operation, the LaunchEngine PM Pack, deployment patterns, and where humans stay in the loop.

Read the guide

Property Management Automation: The Complete Guide

The broader category. Where AI agents fit inside PM automation alongside rule-based logic and workflow orchestration.

Read the guide

How to Implement AI in Property Management Operations

Implementation-focused. Features vs operations, cross-system orchestration, and evaluation rubrics for AI vendors.

Read the guide

Frequently Asked Questions

What AI agents on monday.com are, how the three layers differ, and how to pick the right one.

Software that works inside monday's data model. It triggers on an event (item created, status changed, update posted, cron, external webhook), reads context from workspaces and boards, calls an LLM to make a judgment or generate output, and writes results back to items. Agents can be assigned to items and mentioned in updates like a team member. That is what distinguishes an agent from a chatbot (which waits for the user) or an AI feature inside one app (which does not act on the boards).

monday ships a growing library of pre-built agents. Named ones include Risk Analyzer (project risk detection), Lead Scorer (sales lead scoring), Meeting Summarizer (notes and transcripts), Ticket Assignment (support ticket routing), and Sentiment Detector (feedback and email sentiment). Plus monday sidekick, the general AI assistant embedded across the platform. If a workflow maps to a native agent, that is the fastest path to value.

monday's low-code interface for building custom agents. Three steps: describe what you need, connect knowledge and tools, deploy. Sits under the Agent Factory feature (some customers call the collection of agents they build the Digital Workforce). Right for workflows that need behavior beyond native agents but that fit within the Builder's envelope. No developer required.

Build custom when the workflow needs a specific model, a specific prompt structure, per-run cost telemetry, external system integration beyond monday's built-in connectors, or per-client customization at scale. If the workflow maps cleanly to a native agent or the Builder handles it, start there. Custom is the right layer when the other two do not cover what you need.

Yes. Custom agents that register through the Custom Agents surface get the same account infrastructure as native monday agents. That includes being assigned to items, being mentioned in updates using @, and appearing in item history as the accountable owner of an action.

MCP (Model Context Protocol) is an open protocol for connecting AI agents to tools and data. monday provides two MCP servers. Platform MCP (hosted at mcp.monday.com/mcp) is the primary read/write layer for monday data. Apps MCP (local) covers the app and agent lifecycle: scaffolding, feature creation, deploy, and semantic search across the developer docs. Using MCP lets any MCP-compatible agent runtime work with monday out of the box.

monday grants custom Agents 20M complexity points per minute vs 5M for standard integrations. The typed SDKs (JavaScript and Python) handle rate limits automatically with configurable retry logic. Design your GraphQL queries to be complexity-aware for high-throughput workloads.

OAuth 2.0 Authorization Code Grant with per-user session isolation is the default for user-installed apps. For agent-to-platform calls, a Bearer token with a personal API token works for development. In production, agents typically use their own identity created through the Agent Signup API at /signup-logic/agent-signup/account/v1.

Native and Builder agents consume AI credits from the account's usage dashboard, metered per run. Custom agents on external runtimes have model API costs (Anthropic, OpenAI, or another provider) plus hosting costs (Render, AWS, GCP). Custom agents on monday Code use AI credits for platform-native calls but hit your provider directly for external model calls.

Same app submission flow as any monday marketplace app. Create a developer account, build with the SDKs, test in the sandbox, submit through the Developer Center. Marketplace apps reach 245,000+ monday.com customers. Native monetization handles payment processing for paid apps.

monday supports managed provider integrations that reduce the auth and lifecycle work of standing up a custom agent. The Claude Managed Agent is one such integration: a customer can connect Claude to their monday workspace with less setup than building a runtime from scratch. Right for teams evaluating custom-agent patterns before investing in a full build.

LaunchEngine builds custom AI agents on monday.com for teams whose workflows do not map cleanly to a native agent or the Builder. Full prompt and model control, per-run cost telemetry, and monday-native integration. The PM Pack (Pilot, Echo, Penny, Maxine, Iris, Recap) is live in production for LaunchEngine PMOS customers. Custom builds for workflows outside the pack ship on the same runtime.

About the author

Rob Lowry, Founder of LaunchEngine

Rob has spent 6+ years building on monday.com, from 200-door PM startups to 3,000+ door enterprises. He leads LaunchEngine, which ships custom AI agents and integrations for teams running real workloads on monday.

LinkedIn Team

Building a custom monday.com AI agent?

30-minute intro call. We walk through the workflow, the right layer (native, Builder, or custom), and what a first-agent ungating cadence looks like.

Book an intro call