Atlas
Uses Google Maps as a geo data adapter to resolve geographic data and place information for county maps, with Google geo configured server-side.
Integrates with ChatGPT through the OpenAI Apps SDK, exposing MCP tools and a React widget for interactive county map exploration, place inspection, pinning, and session notes within ChatGPT.
This server does not currently have a deployable release on Glama. Follow the repository's setup instructions to install and configure it for your MCP client, including any required credentials and local dependencies.
Atlas County Maps (ChatGPT App)
Current product (2026-09-10): read-only US Census atlas for ChatGPT. Start at
CHATGPT.md, thenAGENTS.mdanddocs/STATUS.md. Compiled wiki:llm-wiki/wiki/index.md. GitHub map:GITHUB.md.The copy below is historical (voxel plugin era).
docs/NORTH_FACE.mdis not current product law. Do not implement from it.
Product law: AGENTS.md. Historical note only: docs/NORTH_FACE.md.
Atlas is a ChatGPT plugin (Apps SDK / MCP + widget) for high-quality voxel county maps and session notes. The playable proof is Riverside / Eastvale: pan, inspect places, pin, and leave notes that stay in this chat. Other US counties open as honest generated maps with real Census town names. Clawd / Scout / campaign planning is parked for later. Map-first, not a dashboard.
Hero prompts:
Open Riverside County and show the Eastvale map.
Show me Miami-Dade County.
The current engineering route is Alpha-first:
Curated county JSON.
Typed service contracts.
PixiJS
VoxelScenerenderer with SVG fallback.Google Maps behind
GeoDataAdapter, not as the main renderer.Apps SDK/MCP tools as the ChatGPT entrypoint.
This repo already has a lean starter for the OpenAI Apps SDK pattern:
A TypeScript MCP server exposed at
/mcp.A separate React widget bundle rendered inside ChatGPT.
The MCP Apps bridge for iframe communication.
A decoupled data tool and render tool, so the model can reason over structured data before mounting UI.
This repo is intentionally not a generic SaaS shell. The current root server/ and web/ starter is a temporary Apps SDK sandbox; the Atlas monorepo skeleton now lives under apps/ and packages/.
Tooling
Use pnpm. On this machine, global npm fails under walk-up-path, so the scripts are written around pnpm and Corepack.
Required:
Node.js 18 or newer. Current local check found Node
v24.16.0.pnpm. Current local check found pnpm
11.7.0.Git.
No OpenAI API key is required for this local MCP app scaffold. ChatGPT connects to the /mcp endpoint through developer mode and a public HTTPS tunnel or deployment.
Related MCP server: census-mcp
Commands
pnpm install
pnpm env:check
pnpm typecheck:workspaces
pnpm build
pnpm verify:mcp
pnpm devThe dev server listens on:
http://localhost:8787/for a plain health page.http://localhost:8787/previewfor the local widget preview.http://localhost:8787/mcpfor ChatGPT, MCP Inspector, or a tunnel.
pnpm verify:mcp is a client smoke for a running MCP server. By default it
checks http://127.0.0.1:8787/mcp; set ATLAS_MCP_URL to verify another local,
tunnel, or Railway endpoint.
To test with MCP Inspector:
pnpm inspectTo connect from ChatGPT during development, expose the local server with a tunnel such as ngrok:
ngrok http 8787Then create a ChatGPT connector in developer mode with the HTTPS URL ending in /mcp.
Structure
engineering-prompt-pack/ preserved source copy of the engineering prompt pack
prompt-pack/ preserved source copy of the first Atlas prompt pack
prompts/ active Codex prompt sequence
docs/ product and engineering route docs
plugins/ Atlas daemon skill scaffold
data/county_packs/ curated county data
assets/ asset prompts and placeholders
apps/widget/ Vite + React placeholder widget shell
apps/web/ placeholder web shell
packages/config/ shared constants placeholder
packages/core/ core contracts placeholder
packages/geo/ geo adapter placeholder
packages/mcp/ MCP layer placeholder
packages/assets/ asset package placeholder
server/
src/index.ts MCP server, tools, resources, HTTP endpoint
web/
src/ React widget and MCP Apps bridge helpers
dist/ Generated bundle, ignored by Git
scripts/
build-web.mjs esbuild widget build
check-env.mjs local environment sanity check
docs/
apps-sdk-notes.md distilled setup notes from the supplied docsThe official Apps SDK examples repo is cloned under .reference/openai-apps-sdk-examples for inspection and is intentionally ignored by Git.
Current Quest
Current implementation direction is defined by docs/NEXT_QUESTS.md and
artifacts/current-update.json. README is an app overview, not the active
quest authority.
The active ChatGPT county entrypoint is select_county for the
Riverside/Eastvale playable slice:
select_countyask_county_questionrender_voxel_countylookup_world_placespreview_scout_droppreview_campaign_engineget_upgrade_options
The Railway backend is live at https://atlas-backend-production-e6fc.up.railway.app with Google geo configured server-side.
OpenAI Build Week extension
Atlas existed before the OpenAI Build Week submission period. The hackathon entry is the meaningful extension built from July 13 through July 18, 2026, not the earlier prototype.
Work added during the submission period includes:
real 2024 U.S. Census town anchors for all 3,222 supported counties, with 13,797 named places and explicit generated-layout limits outside Riverside;
a hardened seven-tool MCP submission contract, minimized provider results, public support/legal routes, and an exact positive/negative reviewer battery;
stronger ChatGPT emulator fidelity, desktop/mobile visual audits, and a mobile place-sheet interaction fix;
Redis scene-packet worker recovery, health/readiness behavior, release rollback gates, and the first successful production worker deployment; and
a release-gated national Census geography bake for all supported counties, kept behind the
atlasGeoBoardfeature flag.
The main Codex build thread used GPT-5.6 and has session ID
019f687a-82d3-7df0-ac27-5b7ca38b099c. Codex helped audit the existing
product, build and verify the national anchor pipeline, harden worker failure
paths, tighten the MCP contract, run the release and browser proof ladders,
and isolate the submission envelope. The human product decisions stayed
explicit: keep Atlas map-first, submit the focused session-only County Scout,
label generated geography honestly, and leave persistence, payments, and
automated outreach out of the public entry.
The full Devpost copy, demo script, test prompts, and judging handoff are in
docs/DEVPOST_BUILD_WEEK_SUBMISSION.md.
State Model
Server or backend owns business data.
Widget state owns ephemeral view behavior such as selected district/place, stickers, and notes.
Durable cross-session state belongs in a backend storage layer once the product needs it.
The widget can call
ui/update-model-contextwhen UI state should affect the model's next turn.
Avoid putting secrets, tokens, or private data in structuredContent, content, _meta, or widget state.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Related MCP Connectors
Patterns for designing and reviewing AI skills, agents, and multi-agent workflows. Find guidance on context economy, delegation, verification, and tool design; inspect claims, worked examples, maturity labels, and source references. Five read-only tools let agents discover relevant patterns, compare concise cards, read specific sections, and explore relationships. Hosted Streamable HTTP at https://agentic-atlas.dev/mcp/ — no installation, account, or API key required. Browse the atlas at https://agentic-atlas.dev/.
Agent-native MCP server over 49M+ US public and government records, privacy-first, always current.
MCP-native notes and memory for ChatGPT, Claude, and other AI tools.
Rivul is a free, read-only remote MCP server for searching scholarly literature. It runs semantic search over more than 300 million paper records from a dated OpenAlex snapshot, so an agent can find papers by meaning rather than exact keywords, then open the full record for any result. Tools: - search_papers: semantic search with optional filters for publication year range, minimum citation count, open access only, topic and field. Sort by relevance, most cited, newest or oldest. - get_paper_details: the full record for one work by its OpenAlex ID: title, every author, venue, work type, topic, field, publication date, citation count, DOI and open-access links. - usage_status: the workspace's MCP usage for the current day. Results come only from the index, never from a model. Each result carries a stable OpenAlex ID so the source can be checked, and every response states the snapshot date. Useful for literature reviews, scoping reviews, finding highly cited or open-access papers on a topic, citation chasing, and checking that a cited paper exists. Free for every account, with no plans or payment. Sign in with Google through OAuth 2.1 (PKCE, dynamic client registration), or use an MCP key from rivul.ai. Endpoint: https://rivul.ai/mcp (Streamable HTTP). Setup and docs: https://rivul.ai/features/atlas
Related MCP Servers
- AlicenseAqualityDmaintenanceAn MCP server that gives AI agents clean, token-efficient access to US civic & property data — geocoding, census tracts, Opportunity Zones, ACS demographics, and FEMA flood zones — sourced entirely from free federal open data.578 npm1MIT
- FlicenseNot gradedqualityCmaintenanceRemote MCP server exposing US Census (ACS 5-year) and FEMA flood data. Works as a connector in both Claude and ChatGPT.-
- AlicenseNot gradedqualityAmaintenanceEnables AI assistants to query official Census Bureau statistics and metadata, including dataset listings, geography levels, aggregate data, and FIPS code resolution. It also provides prompts for retrieving population data for states, counties, cities, and other geographic areas.MIT
- AlicenseNot gradedqualityAmaintenanceEnables AI assistants to discover and query official Census Bureau datasets, geography levels, aggregate statistics, FIPS codes, and population data through MCP tools and prompts.MIT