Agent Engineering Conference Launched AgentEng: CLI, MCP, A2A and ACP Support for Agentic AI Events

We launched the Agent Engineering Conference to bring together the people building the systems around AI agents. Its Founders Edition takes place at Everyman, Canary Wharf, London, on 16 October 2026. The programme brings founders and engineering leaders into one room for technical talks on context, memory, evaluation, harnesses, inference and coding agents. We intend to continue with future editions in London and San Francisco, alongside the community events already taking place in both cities. As we built the conference website, we gave it a terminal-style interface. Commands appeared above the agenda and speaker sections. They suited the audience, but initially they were part of the presentation. You could read them on the page; you could not install a conference tool and run them yourself.

That prompted a useful question: how should an agent interact with an agent engineering conference? We wanted people to retrieve the programme from a terminal, bring talk abstracts into a coding assistant and query event details through another agent. Once they found an interesting tool, they should also be able to experiment with it in their workspace. Today we are introducing AgentEng: Agent Engineering HQ, our open-source conference and tooling companion. It provides a Python CLI, a Model Context Protocol (MCP) server, an Agent2Agent (A2A) service and experimental Agent Client Protocol (ACP) client support. The source and documentation are available for anyone who wants to use it, study the implementation or contribute.

Watch Demo

AgentEng: Extending the Conference Experience

The conference remains a place to meet builders, hear their technical work and ask questions. AgentEng gives that work a continuing home. You can investigate a topic before attending, bring the programme into your coding workspace, or explore tools afterwards without waiting for the next event. We designed the package around two related resources: the public event catalogue and a directory of agent engineering tools. The event catalogue includes London and San Francisco events, speaker profiles, talks, agendas, venue information, FAQs and official registration links. The directory helps you find frameworks, coding agents, memory systems, evaluation tools, models and infrastructure across the disciplines discussed at the conference. These resources share a typed request and result format. A query made from the CLI, MCP or A2A reaches the same catalogue service. That lets developers choose an interface while retaining the source references and catalogue metadata needed to understand an answer.

CLI: Conference Access from the Terminal

AgentEng is available as a Python package. The website installer includes the CLI and server extras, and can bootstrap uv when needed. If you already use uv, install the package directly with uv tool install 'agenteng[server]'. The documentation covers installation and setup. Run agenteng in a terminal to open the interactive menu, or use individual commands to browse the published programme using the examples below.

The catalogue is bundled with the package. Event and tool lookups work offline, without a provider key or model call. That makes the CLI useful on its own, including when you only need a date, an abstract or a link. The short alias ae is also available. For scripts and agent workflows, place --json before the subcommand. The result includes structured data and supporting sources. When stdout is piped, commands return the shared JSON result automatically. A developer can use that output to build a programme viewer or add conference context to an existing workflow. The installed catalogue contains published event information. Use the official registration link for current availability. The CLI can also query the hosted service through its --remote option.

Agent Engineering Tool Directory

A technical conference should help people navigate the work happening around its subject. AgentEng includes a directory drawn from SuperRadar, bringing together tools, models and infrastructure across the agentic AI ecosystem. It covers twelve disciplines: prompt, context, harness, eval, memory, inference, loop, agentic, code, protocol, graph and search engineering. You can browse a discipline, search for a product, filter by kind or read an individual entry. For example, memory engineering spans several approaches, so the directory gives you a way to inspect the available tools and follow their public links:

Each listing retains source attribution. Readers can see where the information comes from and follow the original project links. Its classifications support discovery; inclusion does not mean that AgentEng has installed, integrated or evaluated every listed product. Readers can follow the linked projects and decide which tools fit their own work. The same directory is available through MCP, A2A and HTTP. An assistant researching a talk can look up relevant tools through the conference service instead of relying entirely on names it recalls from training.

MCP: Integration with Coding Assistants

The MCP server lets a compatible assistant retrieve AgentEng data as a tool. You can keep working in your editor or coding terminal and ask about the programme, a speaker or a tool category. The local server runs over stdio; the hosted server uses Streamable HTTP at a2a.agentengineering.world/mcp/. For clients that accept an mcpServers configuration, a local setup uses the installed executable:

Configuration varies between clients. The commands agenteng connect codex, agenteng connect claude-code and agenteng connect cursor print the relevant setup instructions. They leave your client settings for you to apply. The integration guide covers local and hosted connections. Once connected, questions might include “Show the published London agenda” or “List memory tools.” The server exposes one typed tool named agenteng, with operations for event discovery, talks, speakers, FAQs and the tool directory. A client can request the exact record it needs and retain its source links in the response.

A2A: Access for Independent Agents

MCP gives a client access to tools. A2A gives an independent agent a published interface it can discover and query. AgentEng's hosted service publishes an Agent Card describing its capabilities and connection details. It accepts A2A JSON-RPC requests at a2a.agentengineering.world. A planning assistant can use that interface to retrieve the next published event in either city, inspect a talk or find its official ticket link. The service returns immediate messages for these requests. Developers can get connection examples with agenteng connect a2a. We also expose the same typed requests through HTTP, which makes it easier to start an integration before adding an A2A client:

The conference website's “Talk to AgentEng A2A” panel uses the A2A service too. Public catalogue lookups require no model inference. When configured, conversational chat can use a model with bounded access to the published catalogue and return source references. If that path is unavailable, the service falls back to catalogue answers. The application does not retain those conversations; the data guide explains provider use and hosting logs.

ACP: Conference Context for Coding Sessions

Finding a talk is a starting point. We also wanted attendees to take an idea into a workspace and try it. The experimental agenteng code command acts as an ACP client: it launches a compatible coding agent, attaches the AgentEng MCP server and supplies selected conference or tool records as context. ACP is an optional extra. This installation includes both the server and ACP dependencies:

The example assumes that a Claude ACP adapter is installed and authenticated separately. agenteng code --list reports the supported agents and checks which launchers are on your PATH. AgentEng can also launch other compatible agents, including Codex and Gemini, using their corresponding ACP adapters. You can attach a talk, event, speaker or directory entry with --context. Talk identifiers mentioned in a prompt are detected automatically. A prompt asking for a small demo based on a talk can therefore begin with its published abstract, speaker information and project links, while the attached MCP server supplies additional catalogue lookups.

The terminal displays streamed messages, tool updates and edit diffs. Permission requests received through ACP are presented for your decision and default to rejection. Running the command without a prompt in an interactive terminal opens a continuing chat session. The coding agent uses your own sign-in and model allowance, and works in the selected directory. This support is experimental, and the command options may change. AgentEng is the ACP client here; it does not expose an ACP agent server for editors. The ACP guide documents the supported launchers, context handling, chat commands and permission behaviour.

Conference Information and Event-Day Tools

The event commands include venue details, FAQs, the code of conduct and official ticket links. You can check practical questions before making arrangements. Published answers now include the Founders Edition's travel and accommodation policy and the absence of visa invitation letters. For the conference day, agenteng now and agenteng next use the published timetable. agenteng live provides a full-screen now-and-next board, and agenteng bingo generates a reproducible talk bingo card with printable output options. These features give attendees and organisers small utilities they can inspect and adapt. Agents that discover content through the web also have readable event and tool pages, JSON feeds, event structured data, an llms.txt guide and a sitemap on the hosted service. The aim is to make public information accessible through the interfaces builders already use.

Open Source and Community Contributions

AgentEng's source code is published under Apache-2.0. The repository includes the catalogue service, interface adapters, website exporter, directory importer, tests and deployment instructions. The documentation covers commands, integrations, architecture and data attribution. Conference information comes from the public website, keeping the CLI and hosted agent connected to the programme and details we publish. Developers can inspect how that information is prepared, retrieved and shared across the different interfaces, and use the implementation as a reference for their own projects. Contributions can improve the code, documentation, accessibility or public catalogue. Directory corrections should include primary project links so maintainers can review them. The contributing guide explains the development setup and checks.

Install the CLI, connect an assistant or inspect the repository, and use the published programme as a starting point for your next experiment. Visit agentengineering.world for the London Founders Edition and future announcements. AgentEng is available on PyPI and GitHub, with setup instructions at docs.agentengineering.world.

Our goal is to learn alongside the conferences and community events, giving participants direct experience with the latest developments in agent engineering. We plan to integrate advanced demos and tooling into AgentEng so that people can explore new technologies during the events and continue experimenting afterwards. As we build towards future editions in London and San Francisco, we want the programme and the open-source project to develop together, making AgentEng a useful place to investigate emerging approaches and put ideas from the stage into practice.