@synsci/atlas

Atlas is a graph-based research CLI and bundled skill set. It uses direct REST; Atlas does not expose a public MCP server.

npm i -g @synsci/atlas
atlas login
atlas doctor --format=json
atlas help --format=json
atlas map:view --format=json
atlas nodes:list --format=json

The runtime CLI is direct HTTP. It reads:

• ATLAS_API_KEY
• ATLAS_BASE_URL for non-production APIs
• ATLAS_CLI_CONFIG_PATH for an alternate profile config file

Core Command Surface

Atlas publishes 106 CLI commands plus bundled skills:

• Local auth/config: login, logout, whoami, doctor, config:*, secrets:*
• Nodes and maps: nodes:*, map:*, link:*, access:*, draft:*, labels:*, program:snapshot
• Evidence: evidence:*
• Runs: runs:start, runs:list, runs:stop
• Export/import: history:export, map:export, map:summary:*, map:import
• Ops map export: ops:maps:*
• Integrations: github:*, integrations:*
• Keys: keys:*
• Library/research: library:*, research:*, usage:summary

Run atlas help --format=json for the machine-readable command list and atlas help <command> --schema --format=json for payload shapes.

Agent Onboarding

Atlas uses a browser-first human auth flow and a non-interactive agent runtime. Agents receive Atlas credentials only; library search, grounded Q&A, and research features are available through Atlas commands.

Interactive user

npm i -g @synsci/atlas
atlas login
atlas doctor --format=json
atlas whoami --format=json

atlas login opens the Atlas web login, redeems a one-time exchange token, and writes the local profile config.

Autonomous agent

Paste this prompt into the coding agent. Skills are bundled in the CLI package, so the agent should not install a separate skill package:

You are an agent onboarding me to Atlas by Synthetic Sciences. First ask which email address and organization name I want to use for Atlas, and whether I already have an Atlas API key. If I do, ask me to provide it through a secure secret input or confirm it is already available as ATLAS_API_KEY. If I do not, guide me through Atlas browser login and the web settings page to create one, then wait for me to provide the key.

After credentials are available, install the Atlas CLI with `npm i -g @synsci/atlas@latest`, verify the install with `atlas --version`, run `atlas doctor --format=json`, run `atlas help --format=json`, and confirm the packaged skills are present: atlas, atlas-auto, atlas-lookahead, atlas-prove, atlas-reproduce, atlas-search, atlas-to-graph, and atlas-tree.

Do not configure separate skill downloads or legacy host transports. The skills are bundled inside the Atlas CLI package. Do not print the full API key into shared logs unless I explicitly ask you to reveal it; summarize the key prefix, storage location or environment variable, CLI version, doctor result, and skill count when setup is complete.

Agents should inspect command metadata with atlas help --format=json and per-command schemas with atlas help <command> --schema --format=json.

Examples

atlas nodes:create --payload-json @node.json --format=json
atlas nodes:save --node-id n_abc --payload-json @commit.json --format=json
atlas evidence:add --node-id n_abc --expected-revision 3 \
  --items '[{"local_path":"report.md","artifact_type":"text"}]' \
  --format=json
atlas map:export --node-ids n_abc --format=json
atlas library:search --query "retrieval augmented generation" --format=json
atlas library:add --type repository --repository owner/repo --format=json

Use --yes for destructive commands and --idempotency-key when retrying writes from wrappers or automation.

Skills

The package bundles CLI-only Atlas skills; no separate skill install is needed:

• atlas
• atlas-to-graph
• atlas-reproduce
• atlas-lookahead
• atlas-auto
• atlas-prove
• atlas-tree
• atlas-search

Library indexing, grounded Q&A, and research are exposed as Atlas commands. Users do not provide external search credentials to the CLI.