Contact Us 1-800-596-4880
  1. Homepage
  2. Release Notes

  3. Agent Fabric Release Notes

Agent Fabric Release Notes

Agent Fabric is the AI control plane for AI agents, Model Context Protocol (MCP) servers, and APIs built across platforms. You can see all of your services, let agents take action in your existing tech stack, enforce open policies at runtime, and track and optimize the cost of your AI investments.

October 27, 2026

What’s New

You can now configure parallel tool execution for agent networks in Agent Script (.agent) files. Enable parallel_tool_execution to run independent tool calls planned by a subagent or orchestrator concurrently, which can reduce response time. Set a default in the config block or override the setting for an individual node in its reasoning block. You can also use a runtime expression for a node-level setting.

For more information, see Parallel Tool Execution.

October 22, 2026

What’s New

Features

  • You can now create an A2A bridge for Microsoft Copilot Studio agents to make a non-A2A agent A2A-compliant so that Agent Fabric can discover, govern, and orchestrate it alongside native A2A agents.

    For more information, see Make an Agent A2A-Compliant.

  • You can now monitor traffic through Omni Gateway by applying the Akamai API Security (Omni Gateway) policy to your API instances. This traffic includes:

    • Internal traffic between services (east-west traffic).

    • External traffic to endpoints you can’t register at the Akamai edge, such as MuleSoft-managed CloudHub endpoints.

      For more information, see Monitor Traffic Through Omni Gateway.

  • When you apply an Amazon API Gateway policy from agentic tools through the Platform MCP server, you can now scope it to a path, an HTTP method, or an operation, which combines a path and a method. The agent shows the scopes that the policy supports and asks which scope and target you want. The policy details it returns now include the scope each policy applies to.

Enhancements

  • The Review step of the governance strategy Create Policy wizard now tells you when a policy supports more granular scopes, which you can select in the instance-level policy experience.

  • MuleSoft Agent now returns catalog and governance assets from child business groups for users in the root business group.

  • MuleSoft Agent in Slack and Microsoft Teams now takes fewer steps to answer, so responses arrive faster. In Microsoft Teams, answers now appear as they’re generated instead of all at once.

  • MuleSoft Agent in Slack and Microsoft Teams now supports GraphQL and the additional tools available in the embedded MuleSoft Agent.

  • Third-party libraries are upgraded to address reported security vulnerabilities.

Fixed Issues

The release addresses these enhanced MuleSoft experience issues:

Issue Resolution ID

Agent detail pages no longer show Unknown in the Updated field.

W-23016618

The advanced options for policies that you apply to agents and MCP servers no longer include the option to apply the policy conditionally by HTTP method.

W-23267358

MuleSoft Agent now returns catalog and governance assets from child business groups for users in the root business group.

W-23926623

The See Benchmark view on the Optimizations tab now shows an indicator for additional models.

W-23953679

Users who don’t have access to Cost Management now see a permission message instead of a "No LLM proxies found for the selected filters" message.

W-23983310

Optimization cards now show the See Benchmark link and the Why this policy section, and show potential savings in dollars per month instead of tokens.

W-23998776

Searching model proxies on the Optimizations tab now matches instance IDs in addition to names.

W-24010016

The submit button on policy forms in the Platform MCP server now reads Apply policy instead of Request policy creation.

W-24027218

MuleSoft Agent in Slack and Microsoft Teams now tells you when it can’t find data instead of showing an empty card.

W-24105572

A2A policies now appear in the policy catalog for A2A v1.0 agent instances, so you can apply policies such as A2A Token Based Rate Limit.

W-24119017

The MuleSoft Agent welcome message in Microsoft Teams no longer appears twice when you add the agent to a channel or direct message.

W-24135116

The error message that appears when a user who isn’t an Anypoint Platform administrator grants access permissions is now clearer.

W-24159946

Assets discovered by scanners no longer appear as managed instances with options to apply policies or as deployed to the Production environment.

W-24276575

The Security tab now shows data from all environments instead of only the default environment.

W-24366135

Third-party libraries are updated to resolve reported security vulnerabilities.

W-24505378

October 21, 2026

What’s New

  • You can now deploy your agent network locally and test each broker through an Agent-to-Agent (A2A) chat interface directly from Anypoint Code Builder. Use local deployments to iterate on your agent network without publishing or deploying it.

    For more information, see Deploy and Test Agent Networks Locally.

  • You can now configure OAuth 2.0 client credentials authentication for LLMs in agent networks. For more information, see Agent Network YAML Reference.

  • You can now reuse existing MCP assets from Exchange that use SSE transport.

  • Agent networks now support Agent Script dialect version 1.4.

  • You can now configure authentication for agent cards that use dual egress upstreams. For more information, see Agent Network YAML Reference.

  • This release fixes an issue in which deployments failed when an agent network reused an existing asset from Exchange.

Fixed Issues

Issue Resolution ID

Declared variables are now validated when you build an agent network.

W-24455489

Deployments now preserve the proxy-aware Axios agent.

W-24195003

Builds now fail with error code 1035 when a broker has pushNotifications enabled.

W-23930153

Metadata for version 1.0 agent networks now includes an applications entry.

W-24235374

Agent network builds now detect variable name collisions.

W-24335536

October 14, 2026

What’s New

Features

  • You can now create a ServiceNow AI agent scanner to discover agents from ServiceNow.

  • You can now create a Kong Gateway scanner to discover MCP servers from Kong Gateway.

  • You can now use OAuth 2.0 Authorization Code for Google Vertex AI scanners to create, authorize, and test the provider connection before you continue scanner setup.

    For more information, see Scanner OAuth Authorization.

  • You can now apply open policies at the path, operation, or HTTP method level for providers that require a finer scope than the API instance, such as Amazon API Gateway. When an open policy maps to more than one native policy, the Review step tells you, and the policy list shows the path or operation that each policy applies to.

  • The Platform MCP server now includes tools to apply open policies and to apply and edit native policies from agentic tools, such as Claude and Cursor. A new skill guides agents through applying an open policy.

  • You can now register remote MCP servers that require browser-based OAuth 2.1 Authorization Code with PKCE authentication.

  • You can now publish a custom LLM provider to Anypoint Exchange as an llm-metadata.json classifier asset and select it as the provider for a model proxy route.

    For more information, see Publishing Custom LLM Providers to Anypoint Exchange.

Enhancements

  • Policy configuration forms now support conditional fields and mutually exclusive configuration options, such as the signing key length in JWT Validation and the JWKS source in Apigee Verify JWT.

  • The Configure Policy step of the governance strategy Create Policy wizard now shows the providers that support the selected open policy.

  • Policy validation errors now appear in Spanish when the UI language is Spanish.

  • Dropdowns now support type-ahead search, so you can quickly find items in long lists, such as AWS regions.

  • The Session Token field is now available for Amazon Bedrock routes in dynamic mode, so callers that use short-lived AWS credentials can route requests through a model proxy.

  • MuleSoft Agent in Microsoft Teams now shows the same loading, authentication, and error messages as in Slack, and suggests follow-up actions after each answer.

  • MuleSoft Agent now responds faster in Slack and Microsoft Teams.

  • The MuleSoft bot is now added to a Microsoft Teams channel automatically when you select the channel, and the bot name now includes the region, such as MuleSoft (EU).

  • Instance detail, gateway, and policy template pages load faster.

Changes

  • Universal policies are renamed to open policies, and the Omni & MuleSoft badge for advanced policies is renamed to Omni & Mule Gateway.

  • The Policies tab and activity log no longer appear for MCP server and agent instances that are managed by third-party gateways, such as Azure API Management or Amazon API Gateway, because applying policies isn’t supported for these providers. Previously, the tab showed an Apply Policy button that returned an error.

  • Azure OpenAI no longer appears as a provider option when you create a semantic service, because it isn’t supported.

  • The welcome dialog now reflects the Agent Fabric brand.

Fixed Issues

The release addresses these enhanced MuleSoft experience issues:

Issue Resolution ID

Public MuleSoft catalog assets no longer show a Not Validated conformance badge.

W-22108561

The create_omni_gateway MCP tool now marks runtimeVersion as required, so requests no longer fail with a 400 error.

W-22624722

The Gateways page no longer lists a self-managed gateway that doesn’t exist in Anypoint Platform.

W-22732162

The view_api_instance_policies MCP app now shows the name of the API instance whose policies it displays.

W-22831480

Filters on portfolio pages now narrow results consistently.

W-22848133

Governance strategies created with the prepare_automated_policy_creation MCP tool now appear in your organization’s governance strategies list.

W-22854736

Agent detail pages no longer show an unknown environment.

W-23017333

The list_governance_strategies MCP tool now labels controls as controls instead of guardrails.

W-23059789

MuleSoft Agent in Slack and Microsoft Teams now sends each response as one message instead of several messages.

W-23553477

Creating a Trusted MCP server through MuleSoft Agent in Slack no longer fails at gateway selection.

W-23640564

Installing the Slack app now succeeds for organizations that have an orphaned Slack connection.

W-23756773

Custom asset icons now appear in the portfolio catalog, not only on the asset detail page.

W-23815011

Adding or removing a model proxy route no longer resets the client identifier configuration in the LLM Proxy Core policy, so token tracking continues to work.

W-23824232, W-24138884

The Amount Saved card on the agent dashboard now includes only request compression savings.

W-23943885

The Saved column on the MCP Servers tab of the Optimizations page now shows a trend indicator.

W-23954561

The applied policy count for Payload Optimization now updates after you apply a subpolicy.

W-23955262

The MCP Servers tab on the Optimizations page now paginates results when there are more than 10 MCP servers.

W-23955499

API instance names on Cost Management opportunity cards now appear as links.

W-23974322

Bulk update failure messages for alerts are now shorter.

W-23980698

Selecting the MuleSoft logo in the navigation sidebar now opens the home page.

W-23997715

MCP clients no longer need the full OAuth scope to connect to the MCP server.

W-24006913

You can now use the keyboard to navigate table rows.

W-24015335

A provisioning failure banner no longer appears for a Trusted MCP server that provisioned successfully after a retry.

W-24046223

Editing a disabled open policy on a MuleSoft API instance now shows the saved configuration instead of empty fields.

W-24079553

The consumer endpoint of a model proxy is now always editable, and the port is editable on self-managed gateways and limited to 8081 or 8082 on managed gateways.

W-24087973

The model proxy configuration summary now shows the environment name instead of the environment type, and shows the input format.

W-24104747

Users who aren’t organization administrators can now select a model when configuring a model proxy fallback route.

W-24117065

Saving a model wallet that contains an Amazon Bedrock budget no longer fails with a "model isn’t recognized" error.

W-24119805

The cost and tokens toggle on the Optimizations tab now works without first using it on the Budgets tab.

W-24149250

MCP Bridge now recognizes roles that are assigned through teams when checking permissions.

W-24156034

The Performance page now loads available data when an optional query fails, instead of failing to load entirely.

W-24178872

Users who have the documented API Manager permissions can now create a model proxy without the Organization Administrator role.

W-24194029

Policy configuration fields that accept multiple values no longer split values that contain commas, such as Cedar rules in Attribute-Based Access Control.

W-24198588

Creating an automated policy for A2A v1 agents no longer fails with an "Invalid endpoint types" error.

W-24198627

MuleSoft Agent responses in Microsoft Teams no longer refer to Slack.

W-24215212

Model names now appear when users who have the documented permissions create model wallet budgets or configure model-based routing.

W-24247083

Model detail pages now list model wallets from all environments instead of only the default environment.

W-24254296

Numeric policy fields with fractional steps now show rounded values, such as 0.84, instead of values such as 0.8099999999999999.

W-24255976

Selecting View Optimization in the opportunities carousel now updates the page URL.

W-24269008

Logging in after a session times out no longer fails with an "Invalid OAuth state" error.

W-24276489

You can now edit AWS Signature Injection policies without field errors.

W-24276857

Selecting a model wallet on the Budgets tab now opens the wallet in its own environment.

W-24280601

The Request model override field is now optional for model-based and semantic routing.

W-24313182

An IP address validation error no longer remains after you replace an invalid IP address with a valid one in policy configuration.

W-24359644

Editing a policy on a model proxy route now opens the configuration for that route when more than one route has the same policy.

W-24360920

Model proxies that use the Gemini format now handle streaming responses correctly.

W-24363980

Model proxy descriptions are now limited to 256 characters to match Exchange, so creating a model proxy with a longer description no longer fails.

W-24377785

New model proxy routes no longer show NVIDIA as the provider when you haven’t selected one.

W-24387103

Model proxies with OpenAI and Mistral routes now save without a duplicated routing headers error.

W-24387105

Semantic routing is no longer available for model proxies that use the Anthropic or Gemini input format, because it isn’t supported for these formats.

W-24392230

Creating a model proxy with an Amazon Bedrock OpenAI route no longer fails with a 400 error.

W-24424271

Selecting a self-managed Omni Gateway when editing a model proxy no longer shows an endpoint retrieval error that blocks saving.

W-24440299

August 31, 2026

What’s New

  • You can now deploy agent networks to Anypoint Runtime Fabric (RTF) targets. Your RTF deployment must be on version 3.0.277 or later.

    For more information, see Deploy Agent Network Instances.

  • You can now deploy agent networks that do not include a registry.

  • Agent networks now support the Extended Agent Card from the Agent2Agent (A2A) Protocol specification, enabling richer agent metadata and discovery capabilities.

  • You can now connect the enhanced experience to AWS Secrets Manager, Azure Key Vault, or HashiCorp Vault so API keys and other credentials stay in your secrets manager. MuleSoft stores only metadata, such as secret names and paths, not the secret values. It reads the secret from your vault when a model proxy or other service needs it.

    For more information, see Using Credentials Stored in External Vaults.

  • You can now contain a rogue AI agent before it leaks data, takes unauthorized actions, or runs up costs. Stop that agent’s requests from reaching your models, then restore access if the containment was a false positive.

    For more information, see Detect and Contain Rogue Agents.

  • Cost Management now includes budget management. Set spending budgets for your model wallets, track spend against each budget, and manage budgets across providers and models from the Budgets tab.

    For more information, see Budgets Tab.

  • The Cost Management Optimizations page now includes an Agents tab, so you can review token usage and optimization opportunities for your agents alongside model proxies and MCP servers.

    For more information, see Filter and Search.

  • You can now create, edit, delete, and list model wallets, which allocate token budgets that are granted only when a request’s claims match the wallet’s required claims. Manage model wallets from the UI or through the MuleSoft Agent.

    For more information, see Capping Spend for Callers of Model Proxies.

  • You can now edit instances for model proxies, APIs, agents, and MCP servers from the service detail page or the instance detail page, without switching to Anypoint Platform.

  • Model Proxy now supports semantic caching, TLS security for provider connections, external vault–based provider authentication, and the Model Proxy Request Compression policy.

    For more details about updates to model proxy, see Model Proxy Release Notes.

  • MCP Bridge now exposes tool output schemas as part of the MCP specification, so you can connect servers built with MCP Bridge to Agentforce.

  • These SaaS connectors are now available in the Trusted MCP catalog: Wix, WordPress.com, Hootsuite, Descript, Intercom, Front, and LiveChat.

  • You can now create an A2A bridge to make a non-A2A agent A2A-compliant so that Agent Fabric can discover, govern, and orchestrate it alongside native A2A agents.

    For more information, see Make an Agent A2A-Compliant

  • You can now use OAuth 2.0 Authorization Code for Microsoft Copilot Studio scanners to create, authorize, and test the provider connection before you continue scanner setup.

    For more information, see Microsoft Copilot Studio OAuth Authorization.

  • You can now apply universal policies to your API instances and manage the resulting native policies. A universal policy is defined once and applied as the correct native policy on each supported gateway.

    For more information, see:

  • API scanners now write policies from these provider platforms: Google Apigee, Azure API Management, and Kong Gateway. For these providers, policy results are visible from each discovered API in Portfolio > APIs > Policies.

    For more information, see:

  • You can now track remediation and scan status in the Akamai Security Findings table on the Conformance tab.

    For more information, see Track Remediation and Scan Status.

Fixed Issues

  • Improved performance for agent network operations.

  • Fixed issues with LLM connectivity and Gemini 3 model integration.

August 24, 2026

What’s New

August 6, 2026

What’s New

  • Previously in agent network 2.0, connection and broker IDs were restricted to lowercase letters, numeric characters, and non-trailing underscores. You can now use standard naming conventions, such as upper case letters, dashes, and periods for these IDs.

  • To resolve common deployment errors, see Resolve Agent Network Deployment Errors in Anypoint Code Builder.

July 14, 2026

What’s New

Build agent networks with predictable, governed orchestration using Agent Script — a graph-based language that gives you explicit control over how brokers coordinate agents, tools, and LLMs.

In Agent Network 2.0, agent brokers use guided determinism to separate LLM-powered reasoning from deterministic control flow. With guided determinism you get intelligent decision-making where you need it and guaranteed execution paths everywhere else:

  • Define broker orchestration as a graph of nodes, edges, and triggers in Agent Script (.agent) files. Each node performs a discrete step — either an LLM-powered action like classification or a deterministic action like routing — giving you full visibility and control over execution order.

  • Use probabilistic nodes for complex, open-ended decisions that require nuanced judgment, such as ticket severity classification or intent recognition.

  • Use deterministic nodes to enforce fixed execution paths once a key decision has been made, ensuring reliable, rule-based behavior across operational flows.

  • View a graph-based visual canvas alongside Agent Script code.

  • Organize projects with a dedicated /brokers directory containing .agent files alongside your agent-network.yaml registry and context definitions.

  • Communicate across agents using version 1.0 of the Agent2Agent (A2A) Protocol specification.

  • Create and configure agent networks using natural-language prompts in MuleSoft Vibes, including broker graph design, node configuration, and Exchange asset integration.

  • Build, publish, and deploy agent networks in CI/CD environments using the Anypoint CLI plugin.

For more information, see Building Agent Networks for Agent Fabric.

April 29, 2026

What’s New

These Agent Fabric features are now available in Canada Cloud and Japan Cloud.

Category Feature

Discoverability

Agent Registry: Agent Assets

Agent Registry: MCP Assets (Including Public MCP Assets)

Agent Registry: LLM Assets

Agent Scanners

Governance

Agent, MCP, LLM instances in API Management

A2A, MCP, LLM Omni policies

Managed Omni Gateway in CloudHub 2.0

Managed Omni Gateway in Anypoint Runtime Fabric

Trusted Agent Identity

MCP Bridge

Orchestration

Agent Broker authoring with MuleSoft Vibes (Anypoint Code Builder)

Agent Broker authoring via YAML (Anypoint Code Builder)

Agent Broker deployment to CloudHub 2.0

Agent Broker deployment to Anypoint Runtime Fabric

For details, see Hyperforce Release Notes.

February 17,2026

What’s New

  • Exchange now supports GoDaddy ANS agents. You can now add a scanner to automatically discover and register GoDaddy ANS agents. GoDaddy ANS agents are verified by GoDaddy to be safe and secure. Govern and consume the agents in other applications.

    For more information, see Adding a Scanner for GoDaddy ANS.

February 9, 2026

What’s New

  • Agent Visualizer now includes search and filtering capabilities and MCP server and tool insights.

    For more information, see Agent Visualizer Release Notes.

February 6, 2026

What’s New

January 30, 2026

  • In Anypoint Exchange, you can now add agent scanners to automatically discover and register AI agents across agent ecosystems on external platforms, which include Amazon Bedrock, Google Vertex AI, and Microsoft Azure Copilot. After the agents are registered, govern and consume them in other applications.

    For more information, see Discovering and Cataloging External Services with Scanners.

January 23, 2026

What’s New

  • Anypoint Exchange now includes an option to publish MCPs via URLs instead of uploading a schema file.

    For more information, see Anypoint Exchange Release Notes.

  • You can now create and manage instances for publicly available MCP servers in API Manager.

    For more information, see API Manager Release Notes.

January 15, 2026

What’s New

December 11, 2025

What’s New

November 4, 2025

What’s New

October 3, 2025

What’s New

See Also