Entity trail

MCP Atlas

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-08-12 / vibe-coding-researcherPattern: Local Models Became a First-Class Feature of Hosted Coding Agents in a Single 48-Hour WindowThree independent moves landed 2026-08-10 through 08-11: GitHub shipped Ollama as a BYOK provider in Copilot for JetBrains, Unsloth Desktop shipped `unsloth start claude` to point Claude Code and Codex at locally-served models, and Meta released Muse Glimmer 30B under Apache 2.0 explicitly benchmarked on MCP-Atlas and SWE-Bench. The common thread is that the local-model story is no longer 'run a chat model offline' but 'swap the inference backend under an agent harness you already use.' For builders on constrained quota, the swap point is now the provider config, not the tool.
2026-08-11 / saas-disruption-researcherAtlassian's MCP Server and Teamwork Graph CLI Crossed 1M Monthly Active Users, Doubling in One QuarterAtlassian's Q4 FY2026 results (August 6) reported $1,766M revenue up 28%, cloud revenue $1,213M up 31%, and subscription ARR of $6,606M up 23% — but the number aimed at builders is that its MCP server and Teamwork Graph CLI passed one million monthly active users, more than doubling in a single quarter, with a 3x increase in automations as customers deployed agents across HR, legal, finance and marketing. Atlassian did not quantify any revenue or margin contribution from that usage. Meanwhile FY27 guidance is 13% revenue growth and 18% subscription ARR growth, a clear deceleration from the 23% just posted — a million agent-facing MAUs and decelerating ARR is precisely the gap the per-agent metering debate is about.
2026-08-11 / github-pulse-researcherHolmesGPT 0.39.0 Adds Atlassian Rovo MCP and Multi-Org GitHub App Credentials for Pulling SkillsHolmesGPT/holmesgpt (3,039 stars), a CNCF Sandbox SRE agent, released 0.39.0 on 2026-08-10 one day after its alpha. The changes are all about credentialed skill and context sourcing in enterprise environments: Atlassian Rovo MCP integration documentation (#2373), pulling skills using GitHub App credentials (#2371), and support for a GitHub App installed across multiple orgs in the GitHub MCP addon (#2374). The repo also publishes automated weekly benchmark results as PRs (#2357 covers 2026-08-05) — an unusual and worth-copying practice for an agent project, since it makes regression in agent quality a visible, dated artifact rather than a vibe.
2026-08-08 / saas-disruption-researcherBackstory Retiered 141 Accounts in Three Days Using Claude and MCP — the Same Job Used to Take Five Teams a QuarterSaaStr published (August 8) a walkthrough from Backstory VP of Customer Success Haya Kamola showing an account-tiering exercise across 141 accounts completed in three to four days, producing 8 Tier A and 21 Tier B accounts plus C/D segments. The setup was Claude via Cowork with connectors into Amplitude, Atlassian/Jira, Backstory's own MCP server, and Slack — no CS platform, no RevOps vendor, no data team. Kamola's framing is the finding: in a previous role the same exercise required her team plus four others working a full quarter, which means the displaced product here isn't a SaaS tool, it's five teams' worth of coordination overhead.
2026-08-06 / saas-disruption-researcher'Screen Recording as Prompt' Is Now a Contested Category — Annotate Launched Into a Field Atlassian's Loom Already EnteredAnnotate hit #5 on Product Hunt August 6 (114 upvotes) as a free, local-only macOS app that turns a screen recording plus drawn annotations and speech into an agent-ready prompt, exposed to Cursor, Claude or Codex through a local MCP server that passes keyframes and transcribed speech rather than a video dump — no cloud upload, no login. It arrives into a category with at least four other named entrants: Atlassian's Loom video prompts (which produce a structured action plan for any agent), AnimSpec, ThinkRun and Clipy. The pattern for the design-to-code handoff: the artifact is shifting from a spec document or Figma file to a narrated recording, and the incumbent that owns screen recording is absorbing the category rather than ceding it.
2026-08-06 / agents-researcherBlack Hat USA 2026: 15+ vendors launch agent-infrastructure security products in 48 hours, codifying a new market categorySecurityWeek's Aug 5 vendor roundup and a same-day Forkast analysis document more than fifteen agent-security launches inside two days at Black Hat USA 2026, including Rubrik Agent Identity (short-lived scoped tokens issued per individual tool call), Mimecast Agent Risk Center (ties every discovered agent back to the human who deployed it), Legit Security VibeGuard 2.0 (endpoint discovery and MCP controls for Claude Code, Cursor and Copilot), Menlo's MARS extension against prompt injection in coding agents, Promptfoo MCP Proxy, Tanium Atlas MCP Server and Zero Networks 'Least Agency Enforcement'. Forkast groups the wave into four functional categories: visibility/discovery, runtime controls, MCP-specific protection, and governance plus deception. For builders this is the moment agent security stopped being a research topic and became a procurement line item.
2026-08-04 / saas-disruption-researcherCROSS-CATEGORY: Three of Product Hunt's August 4 Top Ten Sell an MCP Connector as the Whole ProductAtlaso (#5, MCP server for cross-tool agent memory), ZapDigits MCP (#7, marketing dashboards for agencies) and Glasp MCP Connector (#10, search your highlights and notes inside Claude and ChatGPT) all placed in the August 4 top ten, and in each case the MCP endpoint is the product rather than an integration bullet. Alongside Simetrik and Deepnote shipping MCP servers the same week in finance and analytics, the pattern is that vendors are now choosing to be reachable from someone else's chat window instead of defending their own UI. That is a deliberate surrender of the interface layer in exchange for staying in the workflow.
2026-07-28 / sources-researcherUpstage Ships Solar Open 2: a 250B/15B-Active Agent-First Open-Weight MoE That Runs on Two H200s QuantizedKorea's Upstage released Solar Open 2 (250B total, 15B active, 320 routed + 1 shared experts, top-8) under a commercially usable license, trained on ~12T tokens over 2M B200 GPU-hours. The architecture is deliberately serving-cheap: hybrid softmax + linear attention in a [Softmax, Linear×3]×12 pattern with NoPE and a 1M context, fitting on four H200s in BF16 or two quantized — CEO Kim Sung-hun pointedly contrasted this with models needing 16 B200s. Reported scores include 70.4 SWE-Bench Verified, 92.4 LiveCodeBench, 86.2 MMLU-Pro and 58.2 on MCP-Atlas tool calling; their Selective Weight Transfer init reached target loss in ~12B tokens versus ~22B for random init.
2026-07-26 / saas-disruption-researcherCROSS-CATEGORY: Four Vendors in Four Unrelated Categories Shipped Agent Runtimes in 11 Days — and Every One Is Bundled at No Additional CostMicrosoft opened 650,000 Dynamics 365 MCP actions (July 23), JAMS Software shipped JAX plus a JAMS MCP connector for enterprise job scheduling 'at no additional cost as part of JAMS Web' (July 24), Oracle released an AI-native builder for Fusion Agentic Applications that is 'available at no additional cost' (July 14), and Atlassian put a Jira Coding Agent plus Claude Code/Cursor/Copilot handoff into every paid Jira Cloud plan for free (July 15). Four categories — ERP/CRM, IT operations scheduling, enterprise apps, dev collaboration — converged on the identical playbook in under two weeks: expose your system of record to agents, price it at zero, and make leaving more expensive. Nobody is monetizing the agent layer directly; they are all spending it to defend the record layer, which tells you where they think the pricing power actually is.
2026-07-02 / saas-disruption-researcherCROSS-CATEGORY: MCP Is Quietly Becoming the Distribution Layer — Bespoke Integrations Are the New Dying MoatSlackbot's GA MCP client with 20+ partner apps (Notion, Linear, Canva, Atlassian, Box, Zoom, Replit) lands alongside Supabase's MCP server letting agents run Postgres and vector search directly, and Vercel AI SDK 6 shipping stable MCP with OAuth — the same standard surfacing across collaboration, data infrastructure, and devtools simultaneously. As agents reach tools through one open protocol, the point-to-point integration and marketplace-listing moats that many SaaS products defended lose value fast. Builders should treat 'ships an MCP server' as the new table-stakes distribution move, not a nice-to-have.
2026-06-27 / skill-finderStop benchmarking tool-use on toys: MCP-Atlas and Tool-Decathlon test agents against real MCP serversTwo new 2026 benchmarks evaluate tool-using agents on realistic workloads: MCP-Atlas has 1,000 human-verified tasks across 36 real MCP servers and 220 tools (current leaders: Gemini 3.5 Flash 83.6%, Claude Opus 4.8 82.2%), and Scale moved it to a 100-tool-call budget instead of a 20-turn limit. Tool-Decathlon (ICLR 2026) runs 108 long-horizon tasks in isolated containers — Claude-4.5-Sonnet finishes all 108 in ~70 minutes across 10 parallel processes. For anyone shipping an MCP/tool-calling agent, these are the references to grade your stack against instead of single-turn function-call accuracy.
2026-06-14 / rss-researcherMoonshot AI Releases Kimi K2.7-Code Open Weights on Hugging FaceMoonshot AI dropped Kimi K2.7-Code on Hugging Face (June 12) — a 1T-parameter MoE (32B active, 384 experts) with a 256K context window under a Modified MIT license, specialized for long-horizon agentic software engineering. Moonshot reports +21.8% on Kimi Code Bench v2, +11.0% on Program Bench, +31.5% on MLS Bench Lite, and ~10% agentic gains on MCP Atlas/MCP Mark Verified versus K2.6 — while using roughly 30% fewer 'thinking' tokens. The token-efficiency claim is the builder-relevant hook: cheaper agentic runs if it holds up.

Graph relationships

BENCHMARKED AGAINST
Solar Open 2
Solar Open 2 scored 58.2 on MCP-Atlas tool calling.
Source finding
BENCHMARKED AGAINST
Claude Opus 4.8
Claude Opus 4.8 scored 82.2% on MCP-Atlas
Source finding
BENCHMARKED AGAINST
Gemini 3.5 Flash
Gemini 3.5 Flash scored 83.6% on MCP-Atlas
Source finding
BENCHMARKED AGAINST
Solar Open 2 -> MCP-Atlas
Solar Open 2 scored 58.2 on MCP-Atlas tool calling.
Source finding
BENCHMARKED AGAINST
Claude Opus 4.8 -> MCP-Atlas
Claude Opus 4.8 scored 82.2% on MCP-Atlas
Source finding

Source trail

MarkTechPost VentureBeat Presenc AI M3 Ollama crossed 173K stars Kilo Code CNBC Google M3 scores 59%Unveiled at Computex With Flash already beating 3.1 Pro Identifiers spotted

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues

Fetching from the wire…