Skip to main content
Glama
sahil19896

atlas-mcp

by sahil19896

Atlas

A clean, separate rebuild at /home/atlas/atlas. The old /opt/atlas application is not imported, changed, restarted, or used as a database.

Current milestone: Project onboarding and automatic context

Connected servers now have a Projects section. A reviewed local setup command installs bounded source indexing and project-scoped stdio MCP, with automatic context adapters for Codex, Claude Code and Cursor. The hub retrieves published evidence over pinned SSH; managed servers receive no general Atlas credential. This is the source foundation, not complete business-flow understanding or development execution. See onboarding, client coverage and security boundaries.

Each server has a compact identity header and Projects · Overview · Connection navigation. Connected servers open Projects first; Overview holds capacity and measurement freshness, while Connection holds identity, recovery and removal controls. The full-width layout follows the approved Desk theme and motion preferences. See server workspace verification.

Onboarding starts with a project picker: Atlas discovers project folders under /home and supplies the selected name, path and owning development account. Enter a project location retains manual setup for projects elsewhere. New server registration includes the bounded metadata catalog; previously registered links need the displayed discovery update command once. Source indexing and IDE integration still require the reviewed project setup step. See project discovery verification.

The picker treats a Frappe bench as one workspace, with contained apps available on expansion. Separate benches and standalone repositories remain separate choices. Existing app-level catalogs are grouped on the hub without another registration command; local setup verifies the selected bench before indexing its apps. See bench workspace verification.

Automatic connection recovery: after setup, Atlas repairs missing or damaged local key files from a verified surviving copy, replenishes its recovery copy and retries temporary storage/network failures without another server command. The registration, host fingerprint, projects and workloads stay in place. Integrity mismatches and total loss of trusted key material still require a trusted recovery path. See recovery behavior and limits, automatic recovery verification and the original incident record.

Server registration now follows the command-based workflow: generate the on-screen command and a separate ten-minute code, run once on the managed server, and Atlas verifies the identity and connects automatically. Existing unfinished registrations are supported. See command registration verification.

Forget server removes its registration, pairing data, saved health and local SSH identity from Atlas, with durable retries if local cleanup is interrupted. The remote server is unchanged. The same address can be registered again with a fresh identity. See removal verification.

Home follows the owner's old Atlas screenshot: workspace header and status strip, amber Settings icon, expanding sidebar, faint grid, blue pixel animation and lower preferences panel. It uses the approved system typography, the original charcoal palette and launcher proportions, and bare colored navigation icons. Settings and the implemented Servers module are shown. See current Home design verification.

The login design is approved. First-owner setup and login are active at http://192.168.2.95:8080, backed by this rebuild's dedicated PostgreSQL database. HTTP and HTTPS are supported installation modes. See owner login implementation and deployment and current verification. The original login-design acceptance remains historical evidence.

A lightweight, responsive React/TypeScript login interface served by a small Python/FastAPI service on port 8080. All fonts and assets are local. No office renderer, model, database server, or browser worker starts with this page.

The design follows the owner's latest direction: mature system typography, quieter panels, restrained cyan/mint accents and a focused sign-in form. The old pixel-font-heavy interface is not the design template. Its blue cellular animation remains the visual identity; Departure Mono is limited to the Atlas wordmark.

The grid and animation span the full viewport. Content has modest edge gutters and a 1600px reading-width ceiling; the sign-in card is never wider than 430px. Phones put sign-in first and keep counters below it. Desktop layouts are checked for no page scrolling at the tested viewport sizes; small screens retain natural vertical scrolling and accessible controls.

The new sparse 2D animation uses the original cellular shapes, smooth fading and gentle background-pointer interaction. It is bounded to 90 live cells, 180 fading cells, at most 12 frames/second and an eight-million-pixel backing store. It pauses on hidden pages, open dialogs, reduced-motion/high-contrast system preferences, or the visible pause control. Only the pause preference is persisted; no pointer coordinates or form contents are stored. Unavailable canvas falls back to static pixel constellations.

First-owner setup, login, session expiry, logout, Home, and Settings are implemented. An unclaimed installation displays Setup your user after its local setup code is armed. Account creation closes setup; successful login opens Home with Settings as its first module. Settings includes account/connection information, the browser's motion preference, and a guarded factory reset. Servers now supports one-time registration, pinned SSH identities and automatic health checks. Development execution and recovery/MFA remain future milestones. See server registration and security boundaries. See Home, Settings, and reset scope. Functional authentication is not a claim of complete production readiness.

Related MCP server: Atlassian Data Center MCP

Install and operate

Routine source maintenance now has Bench-inspired commands:

./atlas doctor
./atlas build
./atlas restart
./atlas storage
./atlas build --rollback

build compiles and publishes frontend assets using installed tools; it creates no release image and does not restart the service. It retains the current and previous frontend builds, with bounded retention of retired assets. restart verifies the service belongs to this directory and checks application/account-storage health. doctor and storage are read-only. See maintenance behavior and limits and verification.

From this directory, with Python 3.12+, Node.js 24+ and npm on PATH:

./atlas install --port 8080
./atlas status
./atlas logs
./atlas stop
./atlas start

A prebuilt frontend can be installed without Node: ./atlas install --prebuilt --port 8080. This source installer is the first local milestone, not the final signed-release installer discussed in the design.

The current installation is already activated. To obtain your first-owner setup code, run locally:

cd /home/atlas/atlas
./atlas setup-code

Then use Setup your user in the browser. Do not send the code or your password to chat. The code expires after 30 minutes; generating another invalidates the previous code. Once an owner exists, setup cannot be reopened through this command.

For a new installation with installed dependencies and a verified web/dist-auth build, ./atlas enable-login --origin http://HOST:PORT --allow-http provisions a dedicated PostgreSQL container through the existing Docker daemon and activates the staged build. It refuses to overwrite existing authentication settings or storage. The original preview installer refuses to overwrite an activated installation; use a reviewed update for later code changes.

The installer creates its own Python environment and the user service atlas-next.service. It never operates the old system atlas.service. It refuses to take over a busy port or overwrite an unrelated unit. This user's service requires a working user systemd manager; reboot persistence also requires login lingering, already enabled on this host.

The current application listens on IPv4 port 8080; the authentication origin is exactly http://192.168.2.95:8080. No firewall or old reverse-proxy configuration is modified. HTTP does not encrypt passwords or session traffic; its login page states this plainly. HTTPS deployments retain Secure cookies and reject plaintext downgrade attempts. Switching the configured origin invalidates existing sessions.

The dedicated database is container atlas-next-postgres, persistent volume atlas-next-postgres-data, listening only on 127.0.0.1:55438. It has separate migration and limited runtime credentials, generated locally and stored in owner-only files. Do not inspect/share these files or remove the database volume. The container's restart policy is unless-stopped.

Real, minimal public counters

The public endpoint is GET /api/v1/public/overview. It reads only this installation's allowlisted aggregate snapshot at var/public-metrics.json, created once with zero counts during installation. No activity has happened in this new hub, and no old Atlas totals are copied. The file is a small preview storage adapter, not the future application database. There is no public write endpoint.

  • Counts must be nonnegative safe integers. Invalid/missing storage returns unavailable, never fabricated zeros.

  • Estimated context tokens avoided are not billing or cost savings.

  • The browser refreshes once per 30 seconds while visible, stops on hidden tabs, and preserves clearly labelled stale values on connection failures.

  • Snapshot time and last check time are distinct. No fake animated counter changes.

  • Set ATLAS_PUBLIC_METRICS=false in this instance's user-service environment to disable publication. Aggregate volume itself can reveal usage.

Tests

.venv/bin/python -m pip install -r requirements-dev.txt
.venv/bin/python -m pytest tests -q
cd web
ATLAS_TEST_URL=http://127.0.0.1:8089 npm test

Tests use separate synthetic fixtures; do not point credential tests at the real installation. See current verification for the disposable database and HTTP/HTTPS fixtures. Playwright needs Chromium. Either install its browser with npx playwright install chromium or set ATLAS_CHROMIUM_PATH to an existing compatible Chromium executable. Browser evidence is saved under evidence/.

The current host lacks Chromium's desktop libraries. Tests were run in an existing isolated Playwright container without installing host packages. The exact command, results, and screenshots are recorded in evidence/README.md.

Next milestones

  1. Login design approved; PostgreSQL-backed owner setup and login implemented in HTTP and HTTPS modes.

  2. Separately scoped account recovery, password changes, MFA/passkeys, and deployment/backup hardening.

  3. Home module launcher and initial Settings implemented; the pixel office belongs inside Agent Studio, not on every page.

  4. Server registration and bounded automatic health checks implemented; remote development, recovery and additional server operations are separately scoped.

  5. Execution, with isolated runners independently installed and capacity-gated.

  6. React Native/Expo iOS client sharing versioned API contracts, not web DOM components.

The design uses CSS, SVG and one bounded decorative canvas instead of generated raster artwork or a 3D engine. No external fonts, analytics, trackers, or model calls are loaded. Credential requests stay on this installation's configured origin. Automated accessibility checks supplement keyboard and layout tests; they do not replace manual accessibility review.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections. Dates show when Glama detected each change.

No tool schema history has been recorded yet.

Maintenance

ActivityMaintained
ResponsivenessNo issues

Resources

Unclaimed servers have limited discoverability.

Looking for Admin?

If you are the server author, to access and configure the admin panel.

Related MCP Connectors

Related MCP Servers

  • F
    license
    Not graded
    quality
    B
    maintenance
    Atlassian Cloud admin MCP server for project/scheme management, custom field admin, Jira automation CRUD, Assets CMDB, JSM request type/SLA/queue config, Confluence space lifecycle, and org-admin APIs, designed to complement the official Atlassian Rovo MCP.
    -
  • F
    license
    Not graded
    quality
    D
    maintenance
    In-cluster MCP server for read-only diagnostics of Grafana, Prometheus, Alertmanager, and Loki, enabling metric queries, alert listings, and log queries through Grafana datasource proxies with Kubernetes RBAC authentication.
    -

Latest Blog Posts

MCP directory API

We provide all the information about MCP servers via our MCP API.

curl -X GET 'https://glama.ai/api/mcp/v1/servers/sahil19896/atlas-mcp'

If you have feedback or need assistance with the MCP directory API, please join our Discord server