backcontext
v0.3.2
Published
The cloud for your AI agents: shared memory, skills, and secrets synced to Claude Code, Codex, Cursor, and Grok — CLI and MCP server.
Readme
BackContext CLI
The cloud for your AI agents: shared memory (an append-only team journal with hybrid retrieval), synced skills, and policy-gated secrets — connected to Claude Code, Codex, Cursor, and Grok through one CLI and a bundled MCP server.
npm install --global backcontext
backcontext auth login --api-key-stdin # key from Settings → Devices & CLI
backcontext init # journal skill + team skills + MCP for detected toolsThe default service is https://backcontext.com. Self-hosted installations can pass a different URL with --url.
Memory
backcontext retrieve "why did we choose PostgreSQL?"
backcontext context --days 14 --project myapp
backcontext add --title "Decision" <<'ENTRY'
Markdown content…
ENTRYSkills — write agent instructions once, use them in every tool
backcontext skills push ./my-skill # includes SKILL.md and all supporting files
backcontext skills pull # materializes into ~/.claude, ~/.codex, ~/.agents, ~/.cursor, ~/.grok
backcontext skills listDirectory pushes preserve nested scripts, references, assets, binary files and executable flags. Changes to any file create a version. Pull verifies the bundle and protects local edits; use --dry-run to preview and --force to overwrite conflicts. Symbolic links are rejected. Upgrade the server before using bundle push; re-push previously incomplete skills from their original directories.
Secrets — encrypted keys with human-in-the-loop release
echo -n "sk-…" | backcontext secrets set OPENAI_API_KEY --value-stdin --policy ask
backcontext secrets get OPENAI_API_KEY --purpose "calling the API" --wait 300
backcontext run --secrets OPENAI_API_KEY -- npm start
backcontext secrets audit OPENAI_API_KEYask-policy secrets pause until a human approves the request in the dashboard; every access is written to an append-only audit trail.
For agents
Every command supports --json. backcontext mcp runs a stdio MCP server exposing journal_retrieve, journal_context, journal_add, skills_list, skill_get, and secret_get — backcontext init registers it automatically for the tools it detects. Full reference: https://backcontext.com/llms.txt.
Create one machine key per device in the BackContext dashboard. Journals are private by default; teammates read them only after an explicit owner grant. Each key can append only to its owner's journal. Run backcontext --help for the complete command reference.
Upgrading to 0.3.2
npm install --global [email protected]
backcontext --version
backcontext skills doctor
backcontext skills install --backup
backcontext doctorRestart existing MCP sessions to load the updated CLI. This release accepts
--json before or after subcommands, adds feedback show <id>, and relocates
legacy journal-skill backups outside agent discovery without deleting custom files.
Retrieval retries transient failures once, with a 30-second total budget;
retrieve --timeout <milliseconds> or BACKCONTEXT_TIMEOUT_MS customizes that budget.
Historical incomplete skill uploads still need a new push from the original directory.
