Data / MCP / Skills
Data for you. Context for your agent.
Official MCP · AI Engineer
https://ai.engineer/mcpcodex mcp add ai-engineer --url https://ai.engineer/mcpSetup docs
Add the server with the Codex CLI, then start a new agent session.
Codex setup docs ↗Already have MCP servers configured? Keep them and add the ai-engineer entry.
After connecting, enable AI Engineer’s tools and try: “What AI Engineer conferences are coming up?” Your agent chooses and calls the tools for you. No local server required.
Reading with an agent? Request /data with Accept: text/markdown or text/plain, or fetch /data.md. Browser requests keep this interactive view.
Streamable HTTP · Public data + private CODE applications · No API key
Public collections
| Collection | What’s included | JSON | CSV | Explore | Fields & sample |
|---|---|---|---|---|---|
Talks 1,508 rows · ≈4.9M tokens | Recordings, speakers, topics, and summaries | JSON | CSV | Explore | Fields & sampleSampleSample fields and rows follow in the table. |
slug · title · url · videoId · event · durationMs · speakers · topics · summary · slideText · slides Sample row · Selected fields [
{
"title": "Don't Build Agents, Build Skills Instead – Barry Zhang & Mahesh Murag, Anthropic",
"slug": "agent-skills",
"url": "https://ai.engineer/talks/CEvIs9y1uog-agent-skills",
"videoId": "CEvIs9y1uog",
"event": "AI Engineer Code 2025",
"durationMs": 982000
}
] | |||||
Speakers Loading… | Profiles, biographies, affiliations, and talks | Loading… | Loading… | Explore | Fields & sampleSampleSample fields and rows follow in the table. |
slug · name · url · biography · jobTitle · organization · portraitUrl · profiles · affiliations · topics · talkSlugs Loading sample rows… | |||||
Topics Loading… | Ideas connecting the conference archive | Loading… | Loading… | Explore | Fields & sampleSampleSample fields and rows follow in the table. |
slug · name · url · description · talkSlugs Loading sample rows… | |||||
Organizations Loading… | Organizations, their speakers, and talks | Loading… | Loading… | Explore | Fields & sampleSampleSample fields and rows follow in the table. |
slug · name · url · summary · speakerSlugs · code2026SpeakerSlugs · talkSlugs Loading sample rows… | |||||
Chapters Loading… | Chapter titles and timestamps within talks | Loading… | Loading… | Explore | Fields & sampleSampleSample fields and rows follow in the table. |
talkSlug · title · startMs · endMs Loading sample rows… | |||||
Transcript downloads Loading… | Index linking to individual transcript files | Loading… | Loading… | Explore | Fields & sampleSampleSample fields and rows follow in the table. |
talkSlug · title · jsonUrl · csvUrl Loading sample rows… | |||||
How token estimates work
Token counts are estimates, using one token per four serialized JSON characters. Metadata estimates cover each collection download; the transcript estimate covers the linked transcript files rather than only their index and uses published JSON byte sizes where needed. Actual counts vary by model and format. Each download also returns its format-specific estimate in the X-Estimated-Tokens header. CSV exports store lists and nested fields as JSON inside cells.
Loading collection details…
Freshness & coverage
Status JSON ↗Authorized public corpus: Loading coverage…
- Corpus version
- Not recorded
- Source updated
- Not recorded
- Publication recorded
- Not recorded
- Talks with transcripts
- Not recorded
Corpus counts exclude other site records merged into the collections below. Transcript availability does not establish completeness or accuracy. Unknown dates are not inferred from release authorization. Prepared with the catalog release.
Usage, limits & data policy · 600/min global · 60/min per IP
Tool failures return isError: true with an explanation in content, not a success-schema payload. Invalid arguments may return a protocol error or a tool error; inspect both before reading structuredContent. HTTP 429, 529, and 503 are transport failures, not tool outputs.
For filtered talk lookup, transcript, or schedule pagination, pass nextOffset as offset with the returned contentVersion. A null nextOffset marks the end; if content changes, restart pagination.
Usage limits. MCP POST requests share a global limit of 600 per minute, with 60 per minute per IP. This includes tool calls and protocol requests. The shared limit returns HTTP 529; the per-IP limit returns HTTP 429. Respect Retry-After and back off. Need a higher limit? Contact hello@ai.engineer.
Search, then fetch. For a complete speaker or organization inventory, omit the search query and provide the filter, then follow pagination. For topical research, search keywords and fetch supporting passages. Search covers metadata, enrichment, and indexed transcript passages. Summaries are metadata, not quotations. Only transcript excerpts are supported by their recording timestamps; check coverage for unavailable indexing. Fetch surrounding context before drawing conclusions. Transcripts are automated and may contain errors.
Private CODE 2026 applications. Apply through MCP: read requirements, validate answers, then submit after approving the complete application. Applicant records, scores and decisions remain private. This is for CODE attendees, not NYC or speaker proposals. Ticket buying remains a stub; Accelevents ticketing stays on the official conference website. CFPs link to external portals such as Sessionize. MCP cannot purchase tickets, save drafts, or submit a CFP. With the user’s explicit authorization, an agent may use computer use on those official portals to complete a purchase or submission. Confirm the final price or application contents before committing unless already approved, and verify the portal’s confirmation.
Downloads and freshness. The aie://data/catalog resource links to the public collections below. Filtered lookup, transcript, and schedule pagination return content versions so an agent can detect updates. The restricted Hugging Face repository is separate and is not exposed through MCP.
Hugging Face · Restricted access
The research dataset.
Our separate Hugging Face dataset brings together transcripts, segments, topics, entities, summaries, chapters, retrieval examples, and benchmarks in Parquet. It requires authorized access and is not included in the public downloads above.
View on Hugging Face ↗You may need to sign in with an authorized Hugging Face account to view the repository.