Frends vs MuleSoft in 2026: Process Metering, Mule Credits and What They Do to Your Architecture

Overview

Why compare a Finnish iPaaS with Salesforce's integration arm? Because on hybrid projects they collide: both run integrations in their cloud or inside your network, and both now speak MCP. They are very different bets. Frends is a HiQ-owned, .NET-native platform that compiles BPMN-drawn Processes to C#, runs them on Agents hosted by Frends or by you, and counts processes. MuleSoft is Salesforce's Anypoint Platform, running Mule applications on CloudHub 2.0 or your own Kubernetes through Runtime Fabric, and selling annual packages of Mule Credits.

Sources were read on 14 and 15 September 2026 unless dated otherwise; this is a close reading of what each vendor publishes, with our judgement on top.

Company Snapshot

The Real Trade-Off Is What Each Vendor Meters

Feature lists for these two overlap more than either vendor would like. Their meters don't, and the meter is what your architects end up designing around.

Frends counts unique Processes, APIs and Pass-through APIs in use on the last day of the month, once across Development, Test and Production, and calls its execution limits "not a hard-limitation". On Frends, breadth costs money and volume mostly doesn't, so the rational design is fewer, bigger Processes (up to 50 Tasks and 10 triggers each) and polling as often as you like. One Process polling every five seconds runs about 526,000 times a month, more than Growth's whole 0.5M allowance, and Frends' answer is a recommended tier move, not a block.

MuleSoft's pricing page, read 15 September 2026, no longer mentions Mule Flows or Mule Messages. Teams draw Mule Credits from one pool "as they run integrations, manage APIs, automate processes, and use agents", each usage type at its own rate. The page doesn't define the unit, but usage spends credits, so the rational design runs the other way: fewer calls, less polling, traffic consolidated to ride rates that fall as volume grows.

Neither incentive is wrong, and both leak into architecture. Under Frends, watch for monolithic Processes that are cheap to license and painful to debug; under MuleSoft, watch for teams trimming retries and health checks to save credits.

Hybrid Deployment: Both Do It, on Different Terms

Both vendors will run your integrations inside your network. The real questions are which tier unlocks it and which runtime you inherit.

Frends

Hybrid Deployment starts on Business, high-availability Agents on Enterprise and fully air-gapped deployment on Enterprise Infinity (pricing page); Growth v1.0.1 says flatly, "No hybrid-deployment support." Frends 6.3 added Kubernetes Deployment and StatefulSet Agent registration modes, and the Technologies page says horizontal scaling needs a High Availability configuration, which the pricing page lists on Enterprise. If your Kubernetes design assumes scale-out, price Enterprise and have Frends confirm it. Self-hosting also means owning a runtime: on-premise cross-platform Agents need .NET Runtime 10.0.7 or newer (6.3.1), which is trivial for a .NET operations team and new work for anyone else.

MuleSoft

Runtime Fabric deploys Mule applications and API proxies to a Kubernetes cluster you create, configure and manage, on EKS, AKS, GKE, OKE, OpenShift or Rancher. MuleSoft documents the wider menu of Kubernetes targets; Frends makes it clearer which tier buys which deployment right.

API Management: A Tier Feature Versus a Credit Draw

Both will run your gateway. The difference is how the cost lands on your org chart.

Frends

Frends' pricing page lists API Management as a Growth feature and again as an add-on module ("Full API gateway, developer portal, and analytics"), so confirm in writing what your tier includes. For billing, APIs count alongside Processes in the monthly process count, and each pass-through API call is one execution.

MuleSoft

MuleSoft's pricing page (15 September 2026) says teams draw credits from one shared pool "as they run integrations, manage APIs, automate processes, and use agents". That's tidy for a central budget and awkward for chargeback: someone has to decide whose traffic spent the pool.

Connectors: Ignore Both Counts

Frends states "over 300 connectors" on its iPaaS page; MuleSoft's enterprise-integration page offers "hundreds of pre-built connectors" and gives no count. Neither number predicts your project. List your five hardest systems (the ERP, the oldest database, whatever finance refuses to replace) and ask each vendor who maintains that connector and what it supports today.

MCP: MuleSoft Started Earlier, Frends Opened the Whole Platform

MuleSoft has the longer MCP release history. Frends' new launch makes the more interesting governance choice.

Frends

Enterprise MCP, the MCP Trigger that exposes a Process as an MCP tool plus the AI Connector, shipped in Frends 6.3: release notes 4 June 2026, press release 23 June 2026. Frends MCP, launched on 15 September 2026, works at platform level: a developer describes an integration in plain language, Frends returns a Process draft, and the developer "reviews it in the Frends interface and approves it there before it runs anywhere". Every action is "tied to a named user by a personal token that cannot be shared". When an AI client touches a production integration, that's the audit trail you want; issue AI clients personal applications rather than the Tenant-wide ones the docs also describe.

MuleSoft

MuleSoft's MCP Connector, which lets Mule apps act as MCP servers and clients, shipped as 0.1.0-BETA on 15 April 2025, reached 1.0.0 on 26 June 2025 and 1.7.0 on 25 August 2026; its release notes label only 0.1.0-BETA as beta. More than a year of numbered releases counts for something if you're building agent tooling now.

Pricing (as of September 2026)

Headline prices on this pair mislead in opposite directions, so read the meters before the numbers in bold.

Frends (frends.com/pricing)

Four sources give Frends figures, and they are not one price list:

• frends.com/pricing (read 15 September 2026): quote or expert-call buttons, no tier prices; Growth lists 0.5M executions, 100 AI Credits a month and 2 Agents, and one AI Credit costs EUR 0.01 (14 September 2026).
• Azure Marketplace, market FI (read 15 September 2026): EUR 790 (Growth), 2,000 (Business) and 5,000 (Enterprise) a month on one-month, auto-renewing terms, plus per-process meters with no processes included.
• Pricing-page estimator (code read, not run, 14 September 2026): 800 (Growth), 2,000 (Business), 5,000 (Enterprise) and 7,000 ("Enterprise On-Premise"), no currency.
• Vendor-supplied to AutomationAtlas, 11 August 2026: from EUR 800 a month for small businesses, EUR 2,000 for mid-size companies and EUR 5,000 for enterprises; the email described company sizes rather than tier names.

MuleSoft (mulesoft.com/pricing)

MuleSoft's redesigned pricing page, read 15 September 2026, says packages "start at $2,000 per month, billed annually", each with Mule Credits whose rate varies by usage type and "decreases as usage increases". Its buying models are Pre-Purchase ("No minimum credit purchase"), Pre-Commit ("$50K minimum annual commitment") and Unlimited ("Custom fixed price, uncapped credits"). The page gives no effective date and no rate table, so you can't size what the $2,000 buys from public information.

Cost Comparison

Neither pricing page gives a per-unit rate for integration runs, so any comparison that doesn't start from your own workload is a guess.

Editor's Note: After reading both pricing models closely, my advice is to stop comparing headline prices and ask what each meter will do to your architecture. Frends bills by the deployed process: its Azure Marketplace Growth plan includes zero processes and charges EUR 31.40913 for each of the first 25, while runs are "not a hard-limitation". MuleSoft sells Mule Credits, billed annually from $2,000 a month and drawn down by usage. One punishes sprawl; the other punishes chatter. So write one workload down, make both vendors price exactly that in writing, and ask your architects which design each price pushes them towards. One question to settle in writing first: Frends' catalog doesn't say whether its process bands are graduated or volume-priced. — Rafal Fila, ShadowGen

Governance, Admin Controls, and Compliance

The paperwork gap is real, Frends' recovery numbers deserve a second read, and if you govern data tightly, ask where each platform's AI calls actually go.

Frends

Growth (v1.0.1) and Enterprise (v1.0.2) host on Azure in Ireland, the Netherlands and Sweden; Business (v1.0.1) names Ireland and the Netherlands. Frends' compliance documentation, read 15 September 2026, states ISO 27001:2022 certification and links a DNV Cyber attestation file labelled February 2026 but doesn't mention SOC 2, so ask if your questionnaire needs it. Frends MCP access runs on Platform Application tokens with an expiry and scopes, and personal applications "inherit the permissions of the creating user".

Governing data now means governing the models too: companies that keep integration runtimes in their own estate increasingly want the LLMs those integrations call kept there as well. Frends documents a route for that. Its pricing page (read 14 September 2026) offers the "Option to use your own AI models instead of Frends-hosted ones", and a Frends guide read 15 September 2026 connects the AI Connector to a customer-run Ollama server, which with on-premise Agents needs only connectivity "between the local machines". Self-Service Agents require Business or above, and fully air-gapped deployment is listed on Enterprise Infinity. One catch the guide spells out: the AI Connector "does not support additional headers or fields for Ollama", so authentication has to ride in the Service URL, which it says "may not be secure enough to be used over the public internet". Keep that traffic inside your network.

The service descriptions set SLOs of 99.00% (Growth), 99.50% (Business) and 99.99% (Enterprise), and call recovery objectives "not legally binding" while the SLO "does however apply even in disaster scenarios". Look closely at two recovery point objectives: 3.65 hours in Business v1.0.1 and 7.31 hours in Growth v1.0 are exactly 0.5% and 1% of an average 730.5-hour month, the downtime those SLOs allow. Our reading: a downtime budget ended up in a backup field. Growth v1.0.1 now says 15 minutes, while the newest Business description we found still says 3.65 hours. Missed-SLO compensation is 10% of the monthly fee per full percentage point, capped at 30%; by our arithmetic a day-long Business outage already hits that cap, EUR 600 on a EUR 2,000 fee.

MuleSoft

CloudHub 2.0 docs list regions in "North America, South America, the European Union, and Asia-Pacific", and its networking guide deprecates AWS Direct Connect in favour of Transit Gateway attachments. Salesforce's compliance site lists an ISO/IEC 27001:2022 certificate (updated 17 April 2026), a SOC 2 report (updated 10 June 2026) and an ISO 42001:2023 AI management certificate (updated 1 October 2025) for MuleSoft. Runtime Fabric gives MuleSoft its own self-managed runtime; MuleSoft's admin controls, SLA terms and support for self-hosted models were outside the sources read.

Selection Framework: Choose by Team, Meter and Exit Plan

Frends Growth won't suit teams that must run Agents on their own infrastructure (Growth v1.0.1: "No hybrid-deployment support"), and MuleSoft won't suit buyers who want month-to-month terms: all three of its buying models are billed annually.