All analysis was generated autonomously, without human review. Scores are analytical opinions drawn from the cited public sources, without hands-on testing. They are not audits, certifications, investment reports, purchasing advice, or evaluations of quality.
Arcade.dev sells a runtime that decides what an AI agent may do inside a company's systems. It checks each action against the permissions of the person the agent works for and records what it did. Morgan Stanley turns up in two roles. Arcade's careers page describes it as a Fortune 100 company already running the runtime, and a SiliconANGLE article states that Morgan Stanley joined the $60 million funding round as an investor. None of the reviewed sources outside Arcade confirms a customer the company names, and the other adoption figures in them are Arcade's own measurements. Buyers can ask Arcade for a reference among the companies it names, starting with the two its careers page places in the Fortune 100.
| Description | Arcade sells a runtime that sits between AI agents and the business systems they act on, checking each action against the permissions of the user the agent works for, executing it, and recording what happened. | [f1] |
|---|---|---|
| Founded | 2024 | [f2] |
| Funding | $72M total | [f3] |
| Latest funding | Series A, $60M, June 2026 | [f2] |
| Product | What it does |
|---|---|
| Arcade Runtime | Authorizes each AI agent action against the acting user's own permissions, runs it through pre-built Model Context Protocol tools, and records the agent, user and policy decision behind every call. |
AI Defense Matrix
| Govern | Identify | Protect | Detect | Respond | Recover | |
|---|---|---|---|---|---|---|
| AI-Workload Platforms Inference servers, training platforms, vector DB platforms, and the model-loading supply chain. | ||||||
| AI Orchestration Tools Agentic orchestration tools, plus their plugins, skills, hooks, system prompts, scaffolding, harnesses, configuration settings, and MCP clients on user devices. | ||||||
| AI-Generated Code Code produced by AI tools, AI-assisted reviews, AI-generated infrastructure-as-code and tests, and vibe-coded apps that bypass CI/CD. | ||||||
| AI Gateways & Routers MCP proxies and gateways, LLM routers, outbound AI-service traffic, shadow AI egress, and model-registry traffic. | ||||||
| AI Model Model weights, fine-tuning checkpoints, model cards, registries, AIBOM, and the third-party LLMs your enterprise consumes. | ||||||
| Training Data Datasets used for training, fine-tuning, and continued learning. | ||||||
| Runtime AI Data User prompts, inference inputs, RAG content, vector DB content, persistent agent memory, and interaction history. | ||||||
| AI Agent Identities AI agents as non-human principals, plus credentials, keys, permission scopes, service accounts, and delegation chains across agents and tools. |
Arcade Runtime brokers OAuth for AI agents, scopes each action to the acting user's permissions so credentials never reach the agent or the MCP client, filters which tools an agent can see, and keeps a shared registry of the tools and MCP servers in use. It is mapped to the AI Defense Matrix. [f1]
How well the company can compete in its security market, scored across eight dimensions against public evidence.
| Dimension | Score | Rationale |
|---|---|---|
| Problem Clarity How precisely the company defines its problem, with evidence the problem exists at the scale claimed. | 3/5 | Arcade names the buyer, the platform team trying to move agents from demo to production, and the specific blocker, which is that nobody can show an agent acted only within a real user's permissions. A Forbes article describes security teams blocking agents from production, so the pain is corroborated outside the company, but no source in the reviewed record quantifies it. [s1, s5, s11, s6] |
| Capability Depth How specific the technical capabilities are, with evidence beyond marketing claims such as docs and third-party validation. | 4/5 | Documentation covers building tools, running gateways, wiring identity providers and exporting audit logs, and the tool framework is public code under an MIT license that anyone can inspect. Two non-vendor sources corroborate specific capability rather than the pitch: a SiliconANGLE article describes Arcade's URL Elicitation proposal being developed with Anthropic and accepted, and a post by Den Delimarsky lists Arcade.dev among the organizations that worked on the broader authorization specification. [s2, s8, s16, s10, s14] |
| Market Timing Whether the market is ready for this product, with evidence that buyers are actively seeking solutions. | 3/5 | The enabler is dated and credible: the Model Context Protocol acquired a formal authorization specification, published with Anthropic in April 2025 and carried in the protocol's 2025-06-18 revision, with Arcade.dev among the collaborators a post by Den Delimarsky names. Demand evidence stays indirect: one Forbes article on enterprises moving agents out of pilots, capital from Morgan Stanley and Wipro, and a 25-fold rise in tool calls over six months that the same article attributes to Arcade. None of that is analyst category placement, procurement language or budget data. [s15, s14, s9, s11] |
| Team Credibility Demonstrated domain expertise with public signals such as prior exits, publications, and industry recognition. | 4/5 | Alex Salazar co-founded Stormpath, an authentication and authorization API for developers, and TechCrunch reported in 2017 that Okta hired 35 of its 45 employees and licensed the technology rather than buying the company outright. A SiliconANGLE article states that he and chief technology officer Sam Partee started Arcade in 2024 and places Partee in a senior role at Redis beforehand, and two non-vendor sources record standing in the protocol community, SiliconANGLE for the accepted URL Elicitation proposal and Den Delimarsky for the earlier authorization specification. [s12, s9, s10, s14] |
| GTM Proof Evidence of actual traction (customers, revenue signals, partnerships) beyond stated intentions. | 3/5 | Arcade names six companies on its homepage as running production agents and two more on its careers page, its documentation offers deployment through cloud marketplaces alongside its own cloud and Helm self-hosting, and a Forbes article independently records the purchase of the Smithery registry. That is a named commercial record rather than an absent one, which is what keeps it off the floor, but every customer name comes from Arcade's own pages and no reviewed source outside the company corroborates a deployment. [s1, s5, s8, s11] |
| Funding Efficiency Whether funding matches go-to-market ambition, with signs of capital-efficient growth. | 2/5 | The $60 million Series A of June 2026 took total funding to $72 million about two years after founding, and the reviewed sources disclose no revenue, no margin and no customer count. Every named deployment comes from Arcade's own pages, so the raise outruns what the record verifies. [s9, s13, s5, s11] |
| Category Clarity Whether the company creates or fits a recognizable category that buyers can quickly place in their stack. | 3/5 | Buyers can place an agent authorization layer next to the identity and gateway products they already run, and a SiliconANGLE headline uses that plain description. The slot still needs the company's own explanation. Arcade carries two names for one product: an MCP runtime on its homepage, and a secure action layer in the phrase a Forbes article marks as the company's own. The same article calls the category one that is only now becoming a procured layer, with vendors, benchmarks and contracts attached. [s9, s1, s6, s11] |
| Incumbent Defensibility How vulnerable the core value proposition is to absorption as a feature by a platform vendor. | 3/5 | Arcade Runtime handles every agent action, which a Forbes article treats as the position from which a security team can answer which agent did what, on whose behalf, and that is friction a bundled feature would have to reproduce. The reviewed sources do not size that friction, and the authorization specification Arcade builds on was written with Microsoft, Okta/Auth0, Stytch and Descope, so the companies best placed to ship a competing layer helped define the interface. [s2, s11, s14] |
Companies stall AI agents before production because nobody can prove an agent stayed inside a real user's permissions. Arcade's homepage puts the blockers at three: authorization, reliability and governance. Its careers page states the same problem in plainer words, that an agent which can act inside a real company on a user's behalf is the hard part.
The independent record treats this as a security decision rather than a model problem. A Forbes article on Arcade's registry purchase says security teams who have been blocking agents from production now have a shorter list of vendors to evaluate before they stop saying no. That sentence names who does the blocking, which is the buyer-side signal a vendor page cannot supply.
What changed recently is the interface. The Model Context Protocol gained a formal authorization specification, published with Anthropic in April 2025 and carried in the protocol's 2025-06-18 revision, which defines how clients reach restricted servers on a resource owner's behalf, and a post by Den Delimarsky published the day after calls the work a collaboration that included Arcade.dev, Okta/Auth0, Stytch and Descope. A standard authorization flow is what lets a runtime for agent actions be sold as one product instead of built per customer. [s1, s5, s11, s15, s14]
Arcade Runtime decides each action at the moment the agent takes it. The product page states that each action is checked against both what the user may do and what the agent may do, evaluated per action at runtime, with credentials injected at execution and never exposed to the agent or the client. Pre- and post-call hooks let a customer apply its own policy on the same path.
The tool catalog is the other half, and its size depends on what is being counted. Arcade's documentation describes more than 7,500 agent-optimized tools across 81 MCP servers, and on another line calls the same 7,500-plus figure ready-to-use integrations. Its comparison page against Composio then treats Composio's 500-plus as the larger raw count and calls Arcade's catalog smaller. The same word carries two meanings across those pages, so a buyer sizing coverage should ask which unit a quoted number uses.
Deployment and extension are both open. The product page offers a managed cloud, the customer's own AWS, Azure or GCP account, on-premises, and an air-gapped configuration. Custom tools are written in Python against a framework whose repository is public under an MIT license, with evaluation runs against real language models before release. [s2, s8, s7, s16]
Arcade picks its own comparison and it is a tooling company rather than a security one. Its comparison page casts Composio as the breadth option that teams prototype on and itself as the depth option that regulated production deployments land on. That framing puts the contest on tool quality and enterprise authorization rather than on threat coverage.
A Forbes article sets out three positions on the same ground and shows where the pressure comes from. The official MCP registry verifies namespaces and publishes metadata and stops there by design, Docker isolates the process and constrains what it can reach, and Arcade wants to stand where the action is executed, which the article calls the position from which a security team can answer which agent did what, against which system, on whose behalf.
The specification itself is a nearer threat than those rivals. The authorization work Arcade contributed to was written with Microsoft, Okta/Auth0, Stytch and Descope, so several of the companies that could ship a competing runtime already helped define the interface it would implement. Arcade's own Series A post names API gateway vendors and identity vendors as the entrants it expects. [s7, s11, s14, s6]
Every named customer in the reviewed sources comes from Arcade's own pages. The homepage lists LangChain, Snyk, Relevance AI, YUR, Sybill and Eddo Learning as running production agents, and the careers page states that the product is already running inside Fortune 100 companies like Morgan Stanley and OpenTable. None of the reviewed sources outside the company confirms any of those deployments.
Morgan Stanley is worth separating out because it appears on both sides. A SiliconANGLE article states that Morgan Stanley and Wipro joined the $60 million Series A that SYN Ventures led, so the same firm named as a Fortune 100 user is also an investor in the round.
The motion around the names is more checkable. Pricing is public and usage-based, free to start, $25 a month plus $0.10 per authorization event and $0.01 per tool call on the team tier, and custom at the enterprise tier. The documentation offers deployment through cloud marketplaces, the careers page carries an enterprise account executive in New York and forward-deployed engineers in San Francisco and Austin, and a Forbes article records the purchase of the Smithery registry in August 2026 on undisclosed terms. [s1, s5, s9, s3, s8, s11]
The founders have built in this market before. TechCrunch covered Okta's 2017 deal for Stormpath, the authentication API Alex Salazar co-founded, describing it as an acquihire that took 35 of Stormpath's 45 employees plus a licence to the technology rather than a full acquisition. A SiliconANGLE article states that Salazar and Sam Partee founded Arcade in 2024 and places Partee in a senior role at Redis beforehand.
Standing in the protocol community is the second signal and it is corroborated outside the company. A SiliconANGLE article from November 2025 describes URL Elicitation as developed with Anthropic and formalized through an accepted Specification Enhancement Proposal, and a post by Den Delimarsky about publishing the authorization specification lists Arcade.dev among the collaborators.
The company's own claim runs wider than that. Its homepage says the team authored the MCP tool authorization specification and sits on steering committees for MCP security and governance, while its comparison page says it co-authored one extension with Anthropic. The reviewed non-vendor sources support the narrower version. [s12, s9, s10, s14, s1, s7]
The compliance posture is stated by the company and not open to inspection. The homepage and product page say Arcade is SOC 2 compliant with single sign-on, role-based access control and audit logs available, and the comparison page says both Arcade and Composio claim SOC 2 Type II. A probe of trust.arcade.dev on 2026-08-21 reached a live trust center that requires a request for access, while the security subdomain and a nonsense control subdomain did not resolve and the /trust and /security paths returned 404, so no report or certificate is publicly readable at any of the probed addresses.
The controls a security reviewer would ask about are documented in the product itself. Audit logs export to a customer's own monitoring system over OpenTelemetry, each tool call records the agent, the user it acted for, the scope and the policy decision, and deployment can be confined to the customer's cloud, to on-premises hardware, or to an air-gapped environment.
Nothing in the reviewed sources shows an ISO certification, a federal authorization or an independently published audit report. For a runtime whose pitch is that a security team will approve it, an inspectable attestation is the piece that is missing. [s1, s2, s7, s17, s8, s4]
| Company | Relationship | Note |
|---|---|---|
| Composio | competes with | Arcade publishes a comparison page against it and frames the choice as breadth against enterprise authorization depth. |
| Docker | adjacent | A Forbes article sets Arcade's execution-path approach against Docker's process isolation as two answers to the same trust question. |
A closer look at the company's product strategy, measuring how defensible it is against market forces and examining the eight areas behind it.
reinforce or reposition
Arcade.dev's strongest asset is that its runtime handles every agent action, which is what produces the record of what each one did. A Forbes article turns that position into a buyer question. It asks what format the tool definitions, authorization policies and audit records export in, and whether another runtime can read them without a rewrite. The reviewed sources leave that question open, so the mechanism is documented and the cost of leaving Arcade is not. Its other assets are reproducible by a funded rival. The tool catalog is browsable, the registry Arcade bought is public, and the authorization extension it co-authored with Anthropic went into an open specification any competitor can implement. That adds up to a head start rather than a lasting hold.
| Dimension | Score | Rationale |
|---|---|---|
| Value Delivery Does the product sell software as the product, or judgment, trust, or accountability with software as the delivery mechanism. | 1/3 | Arcade delivers software priced per authorization event and tool call, which is the software-product level. A fully managed cloud option removes the infrastructure work, and the enterprise tier attaches a dedicated forward-deployed engineer and service levels, but the customer's team still configures the policies and owns the outcomes. |
| Switching Cost How expensive leaving is for a customer: data portability, integrations, learned workflows, network effects, regulatory data residency. | 2/3 | A Forbes article documents the mechanism, that standardizing on Arcade puts tool definitions, authorization policies and audit records inside one vendor's control plane, and it leaves the exit cost as an open question, telling buyers to ask what format those artifacts export in and whether another runtime can read them without a rewrite. The switching mechanism is documented and the cited record does not size the migration. |
| Compliance Moat Whether certifications, liability acceptance, or audit trails block an easy replacement. | 1/3 | Arcade states SOC 2 compliance and offers single sign-on, role-based access control and audit logs, all of which a funded competitor reaches through ordinary enterprise preparation. A probe on 2026-08-21 found the trust center behind an access-request form and no ISO certification, federal authorization or published report anywhere in the reviewed sources, so nothing here blocks a replacement. |
| Problem Complexity Whether the product requires ML, optimization, real-time systems, or years of specialized expertise. | 3/3 | The runtime evaluates authorization per action against a customer's identity provider while brokering, refreshing and rotating OAuth tokens, and a SiliconANGLE article describes encryption before storage and a salting step for identical plaintext credentials. Parallelized execution with failover and retries, plus evaluation suites run against real language models, is real-time distributed systems work rather than integration plumbing. |
| Buyer Profile Whether buyers are SMB operators, mid-market IT teams, or regulated enterprises and governments with procurement gates. | 2/3 | The documented buyer runs from a free self-service tier and a $25 team tier up to an enterprise tier with air-gapped deployment, a range that reaches well below the regulated-enterprise level and that no customer-mix figure in the record narrows. The one named regulated enterprise, Morgan Stanley, is named as a user only on Arcade's careers page, and it also joined the funding round as an investor, which is thin ground for the higher rung. |
| Layer Whether the product is an end-user application, a platform with application features, or infrastructure other applications depend on. | 3/3 | Arcade Runtime is the layer other software depends on at execution time: agents, frameworks and MCP clients reach business systems through it, and the product page states that the agent never touches raw APIs or secrets. A Forbes article places Arcade where the action is executed rather than in an application a user opens. |
| Proprietary Data, Content, or IP Whether the product accumulates datasets, content licenses, or IP that a rival cannot recreate from scratch. | 1/3 | The cited record evidences no retained asset. The tool catalog is browsable, the Smithery registry Arcade acquired is a public index, the tool framework is public code under an MIT license, and the authorization extension it co-authored went into an open specification. The 25-fold growth in tool-call volume is a usage measurement the company reports about itself, not an accumulated corpus the record shows Arcade keeping. |
Arcade sells to the platform or AI engineering team that owns agent delivery, and the reviewed sources put the security team in the approval path without naming who pays. The homepage promises that agents will sail through security review and the product page contrasts a world where security review kills the project with one where security says yes. Its Series A post names the CIO, the head of AI and the CISO together as the people grappling with the question, without saying which of them holds the line item.
The independent record reads the same way from the other direction. A Forbes article says security teams who have been blocking agents from production now have a shorter list of vendors to evaluate, and separately frames the pitch as one that becomes simple for a CIO. The reviewed sources describe security teams deciding whether agents may run, and none of them records who pays for the runtime.
The segment stretches from free self-service to regulated enterprise. The pricing page opens with a free tier for prototyping and a $25 team tier, and closes with an enterprise tier carrying published service levels, a dedicated account team and air-gapped deployment. The named users run from the six companies the homepage lists to the two the careers page places in the Fortune 100, and beyond that careers-page label the reviewed sources describe the size and segment of none of them.
The capability that carries the product is per-action authorization tied to real identity. Every call is evaluated at runtime against what the acting user may do and what the agent may do, with tokens brokered and refreshed inside the runtime so credentials never reach the agent or the MCP client. A SiliconANGLE article adds detail the marketing pages do not: tokens are encrypted before storage rather than after, and a salting step handles credentials that share plaintext values.
AI is the thing being governed here rather than the method. The tools are built for language-model reasoning, with schemas pruned to what an agent needs and consistent error and pagination handling, and the company runs evaluation suites against real models before a tool ships. That is engineering discipline applied to a reliability problem, not a model the company owns.
Nothing in the reviewed sources describes a model, dataset or corpus that Arcade keeps to itself. The tool catalog is browsable, the framework is public code, and the registry it acquired was already a public index.
Distribution runs through developers first and procurement second. The free tier, the public documentation, the MIT-licensed framework and a job board that is itself an MCP server all aim at engineers who will try the product before anyone signs anything. The paid path then routes through cloud marketplaces, which the documentation offers beside Arcade's own cloud and a Helm self-hosting option.
Enterprise selling is being built out. The careers page carries open roles for an enterprise account executive in New York and forward-deployed engineers in San Francisco and Austin, which is the shape of a company hiring to sell into large accounts and stay to implement. The enterprise tier's dedicated account team and service levels match that intent.
The Smithery purchase is a distribution move as much as a product one. A Forbes article describes an MCP registry with real developer mindshare being bought by the layer that governs execution, which puts Arcade in front of developers at the moment they choose what to run.
Pricing is public and usage-metered. The free tier allows 2,000 authorization events and 2,000 tool calls a month, the team tier adds a platform fee of $25 a month with authorization events at $0.10 and tool calls at $0.01, and the enterprise tier is quoted. Charging per action rather than per seat matches what the product actually does.
The exposure in that model is volume the customer does not control. A Forbes article names it directly as a buyer question, asking what the bill does when one workflow triples its tool calls after a model upgrade, and calls that the scenario that turns a modest platform fee into a line item a finance team wants explained. The enterprise tier answers it with annual bundles and discounted rates as volume grows, which shifts the risk without removing it.
Nothing in the reviewed sources discloses realized revenue, average contract value or gross margin, so the published rate card is all the economics the reviewed sources carry.
Arcade ships software the customer runs, with four deployment shapes rather than one. The product page offers a fully managed cloud, the customer's own AWS, Azure or GCP account, self-managed on-premises, and an air-gapped configuration described as built for classified environments. The documentation adds self-hosting through a Helm chart.
The managed cloud removes the infrastructure work and the self-managed option is offered as full control for the customer. The team tier offers next-business-day email support and the enterprise tier carries 24/7 service levels and a dedicated forward-deployed engineer, which is a services attach on top of the software rather than a service the vendor delivers in place of it.
Reliability engineering is part of the offering rather than an operational afterthought. The product page describes parallelized execution, automatic failover and retries with developer-defined context, and version control for tools and MCP servers so upgrades do not break running agents.
What a security reviewer can verify today is thinner than what the marketing pages assert. Arcade states SOC 2 compliance on its homepage and product page, and its comparison page says both it and Composio claim SOC 2 Type II. A probe on 2026-08-21 found a live trust center at trust.arcade.dev behind an access-request form, no security subdomain, no /trust or /security page, and a nonsense control subdomain that did not resolve, which rules out wildcard DNS as the explanation.
The product-level controls are documented in detail. Audit logs stream to the customer's own monitoring system over OpenTelemetry, every tool call records the agent, the acting user, the scope and the policy decision, role-based access control and single sign-on come with the product, and data residency follows the deployment model the customer picks. The terms of service, effective 15 July 2025, name Arcade AI, Inc. as the contracting party.
No ISO certification, federal authorization or published audit report appears in the reviewed sources. That gap matters more here than for most products, because the pitch Arcade makes is that a security team says yes to the runtime.
Arcade positions itself as neutral infrastructure and the reviewed sources support the description. Its Series A post says the company deploys into a customer's own cloud on AWS, Azure or GCP or on premises, integrates with identity and entitlement systems including Okta, Entra and SailPoint, and works with whichever clients and frameworks a team already uses. The same post frames partnership with model vendors and hyperscalers as the deliberate alternative to competing with them.
Owning the registry complicates that neutrality. A Forbes article calls it the sharper problem, noting that developers adopted Smithery partly on the understanding that it was not owned by a vendor selling the layer underneath, and that publishers now list servers in a catalog controlled by a company whose commercial interest is to run them on its own runtime. The article states that Arcade has not addressed the conflict publicly.
The ecosystem also moves faster than the product can settle. The same article notes a July 2026 protocol release that removed the initialize handshake and session identifiers for remote servers, a breaking change that both registries and runtimes have to absorb.
The founding pair has direct experience of the problem and of the market. Alex Salazar co-founded Stormpath, the authentication API behind Okta's 2017 deal that TechCrunch described as an acquihire of 35 of 45 employees plus a licence to the technology, and Sam Partee came from a senior role at Redis. A SiliconANGLE article puts the founding at 2024.
The wider bench is claimed rather than shown. The homepage says the team comes from Okta, Snowflake, Redis, Airbyte and MongoDB, and the Series A post adds Microsoft to that list, and no employee beyond the two founders and the Smithery co-founder who joined through the acquisition appears in the reviewed sources. The careers page describes a small senior team hiring across engineering, product, marketing and sales.
Protocol standing is the part outsiders can check. A SiliconANGLE article records Arcade's URL Elicitation proposal being accepted into the Model Context Protocol, and a post by Den Delimarsky lists Arcade.dev among the organizations that worked on the broader authorization specification. That is a more specific credential than a list of former employers.
| Id | Source | Tier | Accessed |
|---|---|---|---|
| f1 | Arcade: product page | official | 2026-08-21 |
| f2 | SiliconANGLE: AI agent authorization startup Arcade nabs $60M investment | press | 2026-08-21 |
| f3 | PYMNTS: Arcade books $60 million in funding to keep AI agents accountable | press | 2026-08-21 |
| Id | Source | Tier | Accessed |
|---|---|---|---|
| s1 | Arcade: homepage “The runtime between your agents and every system they need to reach.” | official | 2026-08-21 |
| s2 | Arcade: product page “Compliance SOC 2 compliant SSO and RBAC out of the box Data residency tied to your deployment model Audit logs exportable to your SIEM via OpenTelemetry” | official | 2026-08-21 |
| s3 | Arcade: pricing page “Pay for what your agents actually do, not per seat. Get started for free.” | official | 2026-08-21 |
| s4 | Arcade: terms of service “We/Our/Us: Arcade AI, Inc.” | official | 2026-08-21 |
| s5 | Arcade: careers page “It's already running inside Fortune 100 companies like Morgan Stanley and OpenTable.” | official | 2026-08-21 |
| s6 | Arcade: Series A announcement post “Arcade.dev just raised $60 million in Series A funding. Combined with our seed, that brings our total financing to $72 million.” | official | 2026-08-21 |
| s7 | Arcade: comparison page against Composio “Arcade co-authored the URL Elicitation SEP with Anthropic, now part of the official MCP specification, which standardizes how agents securely authorize with external services.” | official | 2026-08-21 |
| s8 | Arcade: documentation home “Arcade handles OAuth and manages user tokens, enforces your security policies on every action, executes reliably across 7,500+ agent-optimized tools, and governs agents centrally — so your agents take real action in production.” | official | 2026-08-21 |
| s9 | SiliconANGLE: AI agent authorization startup Arcade nabs $60M investment “Arcade AI Inc., the developer of an authorization platform for artificial intelligence agents, today announced that it has raised $60 million in funding.” | press | 2026-08-21 |
| s10 | SiliconANGLE: Arcade.dev and Anthropic advance MCP with new secure authorization flow “Developed in collaboration with Anthropic and formalized through an accepted Specification Enhancement Proposal, the feature gives MCP a standardized and secure way for AI agents to authenticate with real-world services such as Gmail, Slack and Stripe.” | press | 2026-08-21 |
| s11 | Forbes: Arcade acquires Smithery to own the agent tool supply chain “Arcade.dev has acquired Smithery, a leading public registry for Model Context Protocol (MCP) servers, merging its popular discovery platform with Arcade's secure action layer for AI agents.” | press | 2026-08-21 |
| s12 | TechCrunch: Okta acquihires Stormpath, doubles down on identity in apps and APIs “The hiring portion covers 35 of Stormpath's current 45 employees, including co-founders Alex Salazar and Les Hazelwood.” | press | 2026-08-21 |
| s13 | PYMNTS: Arcade books $60 million to keep AI agents accountable “The company's Series A funding round follows a 2025 seed round in which it raised $12 million. It brings the company's total funding to $72 million.” | press | 2026-08-21 |
| s14 | den.dev: The new MCP authorization specification “in collaboration with our friends at Anthropic, we've published a brand-new authorization specification for Model Context Protocol (MCP).” | research | 2026-08-21 |
| s15 | Model Context Protocol: authorization specification, revision 2025-06-18 “The Model Context Protocol provides authorization capabilities at the transport level, enabling MCP clients to make requests to restricted MCP servers on behalf of resource owners. This specification defines the authorization flow for HTTP-based transports.” | other | 2026-08-21 |
| s16 | GitHub: ArcadeAI/arcade-mcp repository “MCP Server Framework and Tool Development library for building custom capabilities into agents.” | other | 2026-08-21 |
| s17 | Attestation probe 2026-08-21: trust.arcade.dev returns an access-request form, security. and a nonsense control subdomain do not resolve, /trust /security 404 | official | 2026-08-21 |
| Id | Source | Tier | Accessed |
|---|---|---|---|
| s1 | Arcade: homepage “The runtime between your agents and every system they need to reach.” | official | 2026-08-21 |
| s2 | Arcade: product page “Compliance SOC 2 compliant SSO and RBAC out of the box Data residency tied to your deployment model Audit logs exportable to your SIEM via OpenTelemetry” | official | 2026-08-21 |
| s3 | Arcade: pricing page “Pay for what your agents actually do, not per seat. Get started for free.” | official | 2026-08-21 |
| s4 | Arcade: terms of service “We/Our/Us: Arcade AI, Inc.” | official | 2026-08-21 |
| s5 | Arcade: careers page “It's already running inside Fortune 100 companies like Morgan Stanley and OpenTable.” | official | 2026-08-21 |
| s6 | Arcade: Series A announcement post “Arcade.dev just raised $60 million in Series A funding. Combined with our seed, that brings our total financing to $72 million.” | official | 2026-08-21 |
| s7 | Arcade: comparison page against Composio “Arcade co-authored the URL Elicitation SEP with Anthropic, now part of the official MCP specification, which standardizes how agents securely authorize with external services.” | official | 2026-08-21 |
| s8 | Arcade: documentation home “Arcade handles OAuth and manages user tokens, enforces your security policies on every action, executes reliably across 7,500+ agent-optimized tools, and governs agents centrally — so your agents take real action in production.” | official | 2026-08-21 |
| s9 | SiliconANGLE: AI agent authorization startup Arcade nabs $60M investment “Arcade AI Inc., the developer of an authorization platform for artificial intelligence agents, today announced that it has raised $60 million in funding.” | press | 2026-08-21 |
| s10 | SiliconANGLE: Arcade.dev and Anthropic advance MCP with new secure authorization flow “Developed in collaboration with Anthropic and formalized through an accepted Specification Enhancement Proposal, the feature gives MCP a standardized and secure way for AI agents to authenticate with real-world services such as Gmail, Slack and Stripe.” | press | 2026-08-21 |
| s11 | Forbes: Arcade acquires Smithery to own the agent tool supply chain “Arcade.dev has acquired Smithery, a leading public registry for Model Context Protocol (MCP) servers, merging its popular discovery platform with Arcade's secure action layer for AI agents.” | press | 2026-08-21 |
| s12 | TechCrunch: Okta acquihires Stormpath, doubles down on identity in apps and APIs “The hiring portion covers 35 of Stormpath's current 45 employees, including co-founders Alex Salazar and Les Hazelwood.” | press | 2026-08-21 |
| s13 | PYMNTS: Arcade books $60 million to keep AI agents accountable “The company's Series A funding round follows a 2025 seed round in which it raised $12 million. It brings the company's total funding to $72 million.” | press | 2026-08-21 |
| s14 | den.dev: The new MCP authorization specification “in collaboration with our friends at Anthropic, we've published a brand-new authorization specification for Model Context Protocol (MCP).” | research | 2026-08-21 |
| s15 | Model Context Protocol: authorization specification, revision 2025-06-18 “The Model Context Protocol provides authorization capabilities at the transport level, enabling MCP clients to make requests to restricted MCP servers on behalf of resource owners. This specification defines the authorization flow for HTTP-based transports.” | other | 2026-08-21 |
| s16 | GitHub: ArcadeAI/arcade-mcp repository “MCP Server Framework and Tool Development library for building custom capabilities into agents.” | other | 2026-08-21 |
| s17 | Attestation probe 2026-08-21: trust.arcade.dev returns an access-request form, security. and a nonsense control subdomain do not resolve, /trust /security 404 | official | 2026-08-21 |
This site is an experimental research aid created by Zeltser Security Corp. All its data gathering and analysis was performed autonomously without human review, and it can contain errors of fact, interpretation, and judgment that a human reviewer might catch.
The analyses are statements of opinion, not statements of fact. Machine analysis produced the scores, summaries, and matrix placements by weighing the public sources each page cites, and reasonable people can weigh the same sources differently. Where a page states a fact, it cites the public source and the date it was checked, and the statement is only as accurate as that source. Unless a profile expressly says otherwise, the analysis involves no hands-on testing and no independent validation of any company's products or services.
Nothing here is professional, security, legal, financial, investment, or purchasing advice, and nothing here is a recommendation to invest in, do business with, or avoid any company. Inclusion of a company is not an endorsement, and absence of a company is not a judgment about it. Reading this site creates no advisory or client relationship. Verify any detail you plan to act on against the vendor's current materials.
The content is provided "as is" and "as available," with all warranties disclaimed, express or implied, including merchantability, fitness for a particular purpose, accuracy, and non-infringement. No entry is warranted to be complete, current, or correct. Companies change, vendors update their claims, sources can be wrong, and automated analysis can misread them.
To the fullest extent permitted by law, the operator, Zeltser Security Corp, is not liable for any damages that arise from using this site or relying on its content, including direct, indirect, incidental, special, and consequential damages and lost profits, even if advised that such damages were possible. If you are dissatisfied with the site or disagree with these terms, your remedy is to stop using it.
Entries link to vendor pages, press coverage, and other external sites that Zeltser Security Corp does not control and is not responsible for. A link is not an affiliation with the destination or an endorsement of it. Product and company names and trademarks are the property of their owners, used here nominatively to identify the companies described. Short quotations from cited sources appear for identification and commentary.
Use, quotation, automated retrieval, and redistribution of the content are governed by the Terms of Use at cybercompanyprofiles.com/terms, which permit personal and internal business use with attribution and prohibit republication and resale.