What it is
Gumloop, made by AgentHub Inc., began as a drag-and-drop builder for AI workflows and now describes itself as a multiplayer AI agent platform. You build agents that connect to company apps such as Slack, Gmail, Salesforce, HubSpot, Notion and GitHub, then run them from chat, Slack or Teams, on a trigger or schedule, or through the REST API. A hosted MCP server lets Claude and other assistants run Gumloop agents and search Company Brain, its store of company documents. The older workflow builder now sits under a "Workflows (Legacy)" label.
There are two plans. Pro starts at $37 a month for 20,000 credits (7,400 plus 12,600 bonus credits) with unlimited agents, seats and teams sharing one credit pool; there is a 14-day trial and no free plan. Enterprise is quoted and adds SAML and SCIM, audit logs, custom data retention, connector policies, model access controls and an optional virtual private cloud.
Credits are indexed at $0.005 each and pay for model tokens, paid tool calls and compute at provider list prices, plus an 8% orchestration fee on every run (16% if you bring your own model keys). Pro credits do not roll over.
Our review
Gumloop's case rests on who can use it. Operations, marketing and sales staff can describe an agent in plain language, connect it to Salesforce or Gmail and put it in Slack, and unlimited seats on Pro mean a whole department shares one $37 plan and one credit pool. The REST API with a published OpenAPI spec and the hosted MCP server give engineers a way in without a separate product.
The pricing rewards staying inside the bundle. Pro's 20,000 credits include 12,600 bonus credits, which works out to roughly $0.00185 a credit, while anything beyond costs $0.005, about 2.7 times more. Pay-as-you-go is off by default, so agents simply stop at the limit until an admin enables it, which protects the budget but can halt a scheduled job mid-month. Bringing your own Anthropic or OpenAI key zeroes the token credits but doubles the orchestration fee to 16% of what the run would have cost at list price, so the saving is smaller than it looks.
Read the privacy policy before connecting a CRM. It says Gumloop may use content from the self-serve tier to train its own models, while Enterprise content is excluded unless the customer asks in writing. Gumloop also says it has agreements with OpenAI and Anthropic that they will not train on data sent through its API. Data is processed in the United States. Certifications live on a trust centre that did not render for this review, so SOC 2 status is unverified here.
The ceilings are practical rather than technical. Pro allows 25 concurrent agent interactions across the organisation and rejects the 26th; Enterprise defaults to 100 and queues instead. For a team running scheduled enrichment across thousands of leads, that difference matters more than any feature on the comparison table. Against n8n, Gumloop trades self-hosting and predictable costs for a gentler builder, which is the right trade for a sales team and the wrong one for a platform group.
Where it earns its place
A four-person revenue operations team at a 200-person SaaS company
Enrich inbound leads and update Salesforce from call transcripts without rep typing
Build a CRM agent triggered by new leads and Gong transcripts. If a month's runs total 60,000 credits including the fee, Pro bills $37 plus 40,000 pay-as-you-go credits at $0.005, about $237. Self-serve content may be used for training, so CRM data argues for Enterprise.
A marketing manager running SEO for a B2B blog
Find decaying pages each month and produce refresh briefs
A scheduled agent pulls Search Console data and drafts a prioritised refresh plan into Google Docs. Light monthly runs fit comfortably inside Pro's 20,000 credits; heavy scraping or web search tools add per-call costs.
A platform engineer exposing internal agents to Claude
Let staff trigger approved Gumloop agents from their assistant
Point Claude at the hosted MCP endpoint over Streamable HTTP with OAuth to list agents and start sessions. Connecting MCP servers inside a private network needs Enterprise's managed tunnels.
Before you commit
- Setup
- Start the 14-day Pro trial, connect apps through OAuth and describe the agent; triggers, schedules and Slack deployment are set in the agent's settings.
- Integrations
- Slack, Teams, Gmail, Salesforce, HubSpot, Notion, Linear, GitHub, Google Sheets and 100+ MCP integrations; REST API at api.gumloop.com/api/v1; hosted MCP server; CLI and Chrome extension.
- Data and privacy
- Self-serve content may train Gumloop's models; Enterprise excluded unless requested in writing; OpenAI and Anthropic contracted not to train; processed in the US; SOC 2 status not verified for this review.
- Cost at scale
- A team whose runs total 100,000 credits a month, orchestration fee included, pays $37 plus 80,000 pay-as-you-go credits at $0.005, about $437; unused credits never roll over.
- Limits
- Pro: 25 concurrent agent interactions, then rejected; 5 concurrent legacy workflow runs; Enterprise: 100 by default and queued; Company Brain uploads capped at 25 files of 25MB per call.
What's good
- Unlimited seats, teams and agents on Pro with one shared credit pool
- Builds agents in plain language for non-engineers
- Hosted MCP server and REST API with an OpenAPI spec
- Pay-as-you-go off by default, so spend stops at the plan limit
- Enterprise adds SAML, SCIM, audit logs and custom retention
What's not
- Self-serve content may be used to train Gumloop's models
- Overage credits cost about 2.7 times the effective included rate
- 8% orchestration fee on every run, 16% with your own keys
- Pro rejects agent runs beyond 25 concurrent interactions
- No free plan; credits do not roll over
Best for
Ops, sales and marketing teams that want to build AI agents on company data without engineers.
Alternatives
Gumloop: common questions
- Is Gumloop worth it?
- Gumloop lets operations and sales staff build useful agents on company data, provided someone reads the credit maths and the self-serve training clause first. We rate it 3.8 out of 5.
- How much does Gumloop cost?
- 14-day Pro trial; Pro from $37/month for 20,000 credits; pay-as-you-go $0.005 per credit (off by default); 8% orchestration fee, 16% with own model keys; Enterprise custom.
- What are the downsides of Gumloop?
- Self-serve content may be used to train Gumloop's models; overage credits cost about 2.7 times the effective included rate; 8% orchestration fee on every run, 16% with your own keys; pro rejects agent runs beyond 25 concurrent interactions; no free plan; credits do not roll over.
- Who is Gumloop best for?
- Ops, sales and marketing teams that want to build AI agents on company data without engineers.
- How much does Gumloop cost for a team or in production?
- A team whose runs total 100,000 credits a month, orchestration fee included, pays $37 plus 80,000 pay-as-you-go credits at $0.005, about $437; unused credits never roll over.
- Is Gumloop safe for company data?
- Self-serve content may train Gumloop's models; Enterprise excluded unless requested in writing; OpenAI and Anthropic contracted not to train; processed in the US; SOC 2 status not verified for this review.
- What are the best Gumloop alternatives?
- The closest alternatives we have reviewed are n8n, Zapier and Relevance AI.
Answers checked Oct 23, 2026.