New Runtime archive

Read-only remote MCP server for the New Runtime AI-engineering newsroom archive: search and fetch evidence-linked signals, analyses, patterns and systems; contextualize a URL; submit sources or corrections.

Hosted MCP Server

npx add-mcp 'https://newruntime.com/mcp'

Installs into Claude Code, Codex, Cursor and more

Documentation

New Runtime agent entrypoint

New Runtime is built to be read by agents: connect, query, verify. A dated, source-linked memory of what is changing in AI engineering; free, read-only, no key.

Last reviewed: 2026-11-13.

Start in 30 seconds

  1. Connect MCP. claude mcp add --transport http newruntime https://newruntime.com/mcp. Stdio-only clients: npx mcp-remote https://newruntime.com/mcp. Claude.ai and Desktop: Settings, Connectors, paste https://newruntime.com/mcp. Cursor, VS Code and Codex snippets: /connect/.
  2. Read the news via API. curl -s https://newruntime.com/posts.json returns JSON. Field Notes newest first: title, description, published_at, topics, and routes.html, routes.markdown and routes.json for every story. Append .md to any story route for its Markdown twin (the full text); search with GET https://newruntime.com/search.json?q=<text>.

Not sure where to start? Read site-index.json (every section, count and page pattern, refreshed on every publication), llms.txt or the developer guide.

I want to…

  • Get the latest news: GET /posts.json · MCP whats_new {"days": 3}. One brief, source-linked Field Note per story, newest first; whats_new also returns signals, analyses and knowledge updates.
  • Pick this week's news for your user (new): GET /for-your-agent.md · MCP get_week_corpus {"corpus": "telegram"}. Follow the instruction: read /weeks/corpus/latest.json, tell your user the token cost, filter one index against what you know about them, and return 5–10 stories with reasons. Their profile never leaves you; an optional anonymous report orders the next site corpus for them.
  • Search the archive: GET /search.json?q=… · MCP search_archive {"query": "codex usage limits"}. Matching Field Notes, Signals, Analysis and knowledge records with title, date, one-line summary and URL; open a hit's .md twin for the full text.
  • Follow a topic: GET /topics.json · MCP get_topic_brief {"topic": "coding-agents"}. Summary, the last fortnight, and the top related patterns and notes for one topic; what_changed {"topic", "since"} gives only the delta.
  • See what changed in a tool: GET /tools/claude-code.md. Claude Code, Codex CLI, Cursor, OpenRouter and Firecrawl release notes, week by week, quoted from the vendor; /tools.json lists every tracked tool.
  • Pick a tool we tested: GET /tested.json. Tools run on the same task and checked by hand: method, results, cost and a verdict. No sponsorship.
  • Research a URL: GET /agent-actions/contextualize.json?url=… · MCP contextualize_url · prompt research_a_url. Which records already cite a release note, repo or post, plus related records and next actions. Free.
  • Verify a claim: GET /source-ledger.json · MCP prompt verify_a_claim {"claim": "…"}. Searches the archive, checks each record's public source_urls and any forecast that covers the claim.
  • See forecasts: GET /forecasts.json · MCP get_forecasts {"status": "open", "topic": "mcp"}. Dated, falsifiable predictions with probability and due date, each tagged with its analysis topics; the hit rate and Brier score fill in as forecasts come due and are graded.
  • Subscribe to changes: Atom + WebSub /changes.xml · MCP what_changed {"since": "2026-10-01"}. Substantive editorial changes pushed over WebSub; per-topic feeds such as /feeds/coding-agents.xml; JSON Feed at /feed.json.
  • Propose a source or correction: Agent inbox /agent-inbox/schema.json · MCP submit_source · submit_correction. Free and moderated: proposals are reviewed weekly and credited in record updates; nothing is published directly.
  • Buy compute per call: GET /agent-payments/payment-capabilities · MCP list_paid_tools. x402, no account: send the request unsigned first; the free preflight returns a 402 quote only when a result is sellable.

Verify factual claims against each record's publishable source_urls; retrieve only the detail JSON or Markdown records you need.

What is published

Since 2026-10-21 every published story is one Field Note at /posts/ with public sources attached; since 2026-10-27 a new note runs 900-1500 words in the sections of an Analysis article (Summary, What happened, What the archive adds, Trend signal, Why it matters, Limits and uncertainty, Reader takeaway, plus optional ones; the archive and trend sections are grounded only in New Runtime's own records), while earlier notes stay 120-250 word briefs. A Signal at /signals/ is a synthesis written only when two or more Field Notes converge on one meaning; records dated earlier are raw observations. Analysis at /analysis/ holds authorial takes from several stories, deep research, retrospectives and the Monday metadox; their forecasts are listed at /forecasts/ and graded as they come due. Field Notes by week are at /weeks/, vendor release notes for the tools builders use at /tools/, what AI companies hire for (aggregate job-posting counts, companies and roles only) at /workforce/, model identifiers seen in public model lists before a launch, with what came true and rumour outcomes by source type, at /radar/, and tools New Runtime ran itself at /tested/. Patterns, Shifts, Deltas and the Graveyard are the trend flow, revised every Monday and read together at /trends/. Older records keep their routes and are not rewritten.

Route reference

Start here

  • Site index (JSON): Every section with its index, page pattern, record count and route templates, under 10 KB. Read first.
  • Search (JSON): Free GET search over the archive (q, limit, kind, since); same data as MCP search_archive.
  • Agent map (Text): Compact route map for a small context budget.
  • Full agent guide (Text): Expanded reading order, citation boundaries, and collection guidance.
  • This page as Markdown (Markdown): Quick starts, task chooser and route reference for a context window.
  • Developer guide (HTML): REST and MCP quickstarts, errors, versions, deprecation, and enforced limits.
  • Connect MCP (HTML): One-line setup of the hosted read-only MCP server for Claude, Cursor, VS Code, Codex and stdio clients.
  • Week corpus for your agent (Markdown): Instruction for a reader's agent: this week's news as plain files (/weeks/corpus/latest.json), picked against the reader's own profile; optional relevance report.
  • OpenAPI (JSON): Typed contract for the public JSON and MCP API.
  • Route catalog (JSON): Complete HTML, Markdown, JSON, text, and XML route catalog.
  • MCP server card (JSON): No-auth Streamable HTTP discovery contract: resources, read-only tools, moderated submission tools.
  • MCP manifest (JSON): Read-only resource catalog and hosted Streamable HTTP contract; no tool edits or publishes records, no approvals or deploy actions.
  • API Catalog (JSON): RFC 9727 linkset connecting the public OpenAPI, MCP, feeds, and discovery surfaces.
  • Design system (Markdown): Paper-and-ink visual system in DESIGN.md format: tokens, identity choices, reading modes, agent prompt guide.
  • Design tokens (JSON): W3C DTCG export of the DESIGN.md colour, typography, spacing and radius tokens.

Content

  • Field Notes (JSON): One brief, source-linked note per story, newest first, with HTML, Markdown and JSON detail routes.
  • Signals (JSON): Syntheses of several Field Notes since 2026-10-21; earlier raw observations with novelty_label and what_changed.
  • Analysis (JSON): Authorial takes, deep research, retrospectives and the Monday metadox; complete list, 25-row pages at /analysis/page/{n}.json.
  • Topics (JSON): Topic hubs with a summary and the records filed under each; latest_live lists the last fortnight.
  • Tool changes (JSON): Vendor release notes by week for the tools builders use; one tool at /tools/{tool}.md or .json.
  • What we tested (JSON): Tools run on the same task and checked by hand, with verdict and cost basis; Markdown /tested.md.
  • Workforce (JSON): What AI companies hire for: aggregate counts of open job postings by role family, tools, skills, titles and pay; pages /workforce/ and /workforce/{company}/.
  • Pre-release radar (JSON): Model identifiers seen in public SDKs, model catalogues, cloud docs and inference-engine code before a launch, confirmed launches with lead days, rumour outcomes and a scorecard by source type; page /radar/.
  • Learn (JSON): AI-engineering reference guides with a weekly fresh-evidence section.
  • Courses (JSON): Curated outside courses on LLMs, prompting, agents and evals: links, levels, hours, cost; verified quarterly.
  • Systems (JSON): Dossiers with per-system release timeline routes.
  • Source ledger (JSON): Public evidence map and record backlinks.
  • GitHub radar dataset (JSON): Versioned authority asset with public GitHub provenance.
  • Change feed (XML): Atom + WebSub feed for substantive editorial changes.
  • JSON Feed (JSON): JSON Feed 1.1 twin for recent Field Notes.
  • RSS (XML): RSS 2.0 feed of recent Field Notes.
  • Complete sitemap (XML): Every public URL: Field Notes (/sitemap-posts.xml), signals (/sitemap-signals.xml) and machine twins (/sitemap-agents.xml) included.
  • Google sitemap (XML): The robots.txt sitemap: only the pages Google indexes (no Field Notes, signals or twins).

Trends & forecasts

  • Trends hub (HTML): Shifts, deltas, graveyard and patterns read together; JSON per collection.
  • Week in review (JSON): Field Notes grouped by ISO week and topic; one week at /weeks/{week}.json or .md (Markdown index /weeks.md).
  • New Runtime data (JSON): Measured company evidence: hiring by department, pricing and docs page changes, open-source adoption; pages /companies/ and /noticed/.
  • Patterns (JSON): Evidence-linked trend hypotheses and revision conditions.
  • Tectonic shifts (JSON): Longer-lived structural interpretation.
  • Capability deltas (JSON): What became possible, with verification and buildability.
  • Interface graveyard (JSON): Actions and interfaces becoming optional.
  • Projects (JSON): Forkable implementation records with proof outputs and replay commands.
  • Forecast ledger (JSON): Dated forecasts with topics, graded when due; the scorecard stays empty until the first is graded (MCP get_forecasts).
  • Agent tasks (JSON): Open verification tasks for stale or uncertain patterns and shifts (MCP list_agent_tasks).

Telemetry & discovery

  • Machine consumption (JSON): Privacy-safe aggregate of crawler and machine-route readership.
  • Machine Discovery Lab (HTML): Crawler identity confidence, graph coverage, latency, and pre-registered discovery experiments.
  • Visitor dossiers (JSON): One dossier per crawler and agent family: operator, purpose, robots.txt, verification and observed baseline.
  • Live bot traffic (JSON): Last-24h aggregate refreshed every 30 minutes; no IPs or raw log lines.
  • Crawler journeys (JSON): Per-family material depth, entrypoints and transitions without IPs or session identifiers.
  • Discovery graph (JSON): Sharded physical-link and editorial-knowledge graph for bounded navigation.
  • Page readership schema (JSON): Contract for per-page machine readership shards.
  • Presence (JSON): Privacy-safe distribution baseline and attributable-return evidence.

Actions & commerce

  • Contextualize a URL (JSON): Free: records that cite an outside URL, related patterns and next actions.
  • Agent inbox schema (JSON): Free, moderated source, correction, usage-receipt, and week-corpus relevance-report submissions; reviewed weekly, never published directly.
  • Paid agent products (JSON): Live product IDs, prices, accepted x402 rails, free preflight, and settled-result replay policy.

Owner profile

Paid compute

The live payment capabilities document is the source of truth for product IDs, prices, networks, assets, preflight, finality, and replay. Send an unsigned request before authorizing any x402 payment. Fixed and ERC-20/SPL routes use x402 exact settlement. Published records remain free. A paid POST computes a derived response but cannot mutate editorial or operational state.

  • For EURC, WETH, cbBTC, or wSOL, use the candidate payment_route marked settlement_enabled=true. Never sign the sibling /quote-preview route.
  • For native BNB, ETH, AVAX, POL, or SOL, use the exact direct_native_payment.candidates[].payment_route: preflight, broadcast the exact transfer with its declared data/memo binding, then submit that transaction identifier. Do not send an unquoted transfer.

Boundary

Agents may cite public source routes and public record pages. Do not infer or disclose private discovery feeds, inbox rules, credentials, handoff files, unpublished drafts, or Telegram operations from this site. External systems cannot modify New Runtime. Paid compute routes return derived results; content, publication, approval, and deploy actions stay inside owner-approved internal workflows.

Public routes expose published records, publishable evidence, and explicitly advertised paid compute products only. They do not expose private discovery feeds, inbox rules, credentials, drafts, private Telegram operations, approval actions, deployment, administration, or site mutations.