Agentic AI Foundation launches MCPA certification for MCP builders
David Soria Parra's open integration protocol now has an exam covering architecture, execution, permissions, and trust boundaries.
By RuntimeWire Staff · Published
Primary source: PR Newswire
Why it matters
MCPA gives AAIF influence over how employers define MCP competence, especially around permissions and security, while giving developers a portable, vendor-neutral credential tied to the open specification.
David Soria Parra, the Anthropic engineer who co-created the Model Context Protocol, is helping turn MCP from a fast-moving open standard into a skill employers can put on a job description.
The Agentic AI Foundation announced the Model Context Protocol Associate certification on September 14th, giving developers a vendor-neutral exam covering MCP architecture, implementation, and security.
AAIF calls the certification the first official MCP credential. The supplied reporting does not establish an exhaustive comparison with every MCP training credential, but MCPA is AAIF's first certification and is tied directly to the MCP specification released on July 28th, 2026.
Soria Parra helped create MCP at Anthropic in 2024 to replace one-off integrations between AI applications and external tools, services, and data sources. MCP applications use a shared protocol in which hosts connect through clients to servers that expose tools. The MCP architecture defines how those components communicate.
"We built MCP so developers could rely on one open protocol instead of writing custom integrations for every system," Soria Parra said in AAIF's certification announcement.
Before co-creating MCP, Soria Parra worked at Facebook from 2013 to 2023, eventually leading developer-infrastructure, static-analysis, and simulation-testing organizations. His biography says he also worked on PHP as a core developer responsible for the PHP 5.4 and 5.5 releases, contributed to Git and Mercurial, and joined Anthropic in April 2024 after serving as an entrepreneur in residence at Sutter Hill Ventures.
That background explains the protocol's bias toward shared plumbing. MCP treats agent connectivity as an infrastructure problem: define a common message format and component model so developers can spend less time rebuilding connectors for every combination of model and service.
An exam built around the risky parts
The multiple-choice, online proctored exam allocates 26% of its questions to interactions and execution, the largest of its five sections. Security and governance account for another 24%, followed by use cases and the surrounding software at 20%, MCP fundamentals at 16%, and architecture and components at 14%.
Candidates are expected to understand hosts, clients, servers, tools, resources, prompts, transports, message flows, error handling, permissions, consent, auditability, and risk controls. AAIF recommends familiarity with JSON-RPC, language-model APIs, OAuth 2.1, token handling, agent patterns, and MCP server manifests, although it lists no formal prerequisite.
MCP implementations can expose data and executable tools to models, making permissions, consent, auditability, and trust boundaries important parts of deployment. The July 28th, 2026 specification is the revision the MCPA uses as its reference point.
The NSA's May 20th, 2026 guidance adds institutional weight to the certification's emphasis on permissions, trust boundaries, and implementation security. The agency identifies serialization risks, implicit trust relationships, dynamic tool invocation, context sharing, and unverified task propagation across agentic systems.
A certification can establish that an engineer knows the terminology and expected controls. It does not establish that the engineer can operate a secure production deployment. MCPA is a multiple-choice knowledge exam, rather than a performance test requiring candidates to build, debug, or secure a working MCP system.
Angie Jones turns adoption into a teaching problem
The other central figure in the launch is Angie Jones, AAIF's vice president of developer experience. Jones has spent much of her career translating software practices into training: she worked as an engineer and master inventor at IBM, later held roles at Twitter and Applitools, and created Test Automation University before leading developer relations for Block, the payments and financial-services company, on its open-source and AI work.
Jones is applying the same education playbook to agent infrastructure. Her stated case for MCPA is straightforward: employers need a shared benchmark as MCP appears in AI engineering, platform engineering, and governance roles.
The exam also gives AAIF a way to shape what baseline competence means before vendor-specific implementations fragment the hiring market. Its weighting makes that definition clear. Exactly half the exam covers execution plus security and governance, placing permissions, consent, and operational behavior alongside protocol mechanics.
MCP gets an institution around the protocol
AAIF is less than a year old. Anthropic, Block, and OpenAI established it under the Linux Foundation on December 9th, 2025, with initial support from Amazon Web Services, Bloomberg, Cloudflare, Google, and Microsoft. OpenAI's launch announcement described AAIF as a Linux Foundation initiative founded to support open agentic infrastructure.
The foundation now houses MCP alongside Agent2Agent, AGENTS.md, goose, agentgateway, and Agent Router. Its thesis is that agentic AI will need the same combination of shared protocols, open-source implementations, governance, training, and professional credentials that formed around earlier infrastructure markets.
AAIF is launching the certification as it gathers builders for AGNTCon and MCPCon Europe in Amsterdam on September 17th-18th. Conference attendees receive a 20% discount, as do registrants for the North American event in San Jose on October 22nd-23rd.
The timing serves both sides of the protocol market. Developers get a vendor-neutral credential designed to travel across tools, employers, and industries, while AAIF gets a mechanism for promoting its preferred implementation and security practices. Employers get a screening signal, although the value of that signal will depend on whether hiring managers recognize it and whether certified candidates perform better on real systems.
AAIF says monthly downloads across MCP's Tier 1 software development kits are approaching 500 million, with the TypeScript and Python SDKs passing 1 billion cumulative downloads combined. It also says MCP tool calls from ChatGPT users reached 98 times their January level by August. Those figures come from AAIF and measure software distribution or usage activity, rather than the number of production deployments or trained engineers.