Terminal — CLI
npm install and your agent signs itself up. Works in any shell, any CI run.
npm install -g @hebrah/cli hebrah signup --name "my-agent" --email you@example.com # → hb_conn_… key saved to ~/.hebrah/credentials (mode 0600)
Give your agent private data access.
One wallet, three ways to connect, 90-second signup. Scoped connections to 12 demo targets and 8 live account connectors — and every query leaves a hash-chained audit your customers can verify.
Give this to your agent — it takes it from there.
Explore agent map nodes
Your agent
One scoped agent key opens connections to every source it needs — Postgres, Stripe, Notion, Google Drive, Slack — each with its own scopes, TTL, and audit chain.
Three ways to connect
Same registry, same wallet, whichever your agent speaks.
Hosted MCP
Point any MCP host at the hosted endpoint. The agent never sees a data-source credential.
{
"mcpServers": {
"hebrah": {
"url": "http://localhost:3021/mcp",
"headers": { "Authorization": "Bearer hb_conn_YOUR_KEY" }
}
}
}HTTP — headless signup
One POST against the agent API, no dashboard. Same registry, same $1 trial, same wallet.
curl -X POST https://api.hebrah.com/v1/agent/account \
-H "Content-Type: application/json" \
-d '{"orgName":"my-agent","agentName":"my-agent","headless":true}'
# → 201 {"apiKey":"hb_conn_…","orgId":"…","trial":{…}}The 90-second workflow
This is the exact flow from /SKILL.md — your agent can run it unattended.
1
Create the account
No card required. $1 credit, 100 queries, 5 MB egress, 7-day trial.
POST /v1/agent/account
{"orgName":"my-agent","agentName":"my-agent","headless":true}2
Discover targets
20 connector targets: 12 demo across 6 packs (Postgres, SQLite, SFTP, Stripe, FHIR, Notion, Slack, Linear, and more) plus 8 live account connectors (your own Stripe, Postgres, Notion, Slack…). Scope grammar varies per target: inspect before connecting.
GET /v1/connections/targets # → 20 targets (12 demo + 8 live) · packs: audit, developer, fintech, # healthcare, manufacturing, workspace
3
Open a scoped connection
Least privilege by default: name the tables and period. TTL bounds the blast radius.
POST /v1/connections
{"target_id":"demo-sqlite-research",
"scopes":["table:articles","period:2025-Q4"],
"tier":"container","ttl_seconds":3600}
# → 201 {"connectionId":"…","audit_genesis_hash":"…"}4
Query through the connection
Server-side scope rewrite: SELECT * returns only in-scope rows. Cost per query comes back on every response.
POST /v1/connections/{id}/query
{"sql":"SELECT * FROM articles LIMIT 3"}
# → 200 {"rows":[…],"cost_cents":2,
# "scope_applied":{"tables":["articles"]}}5
Export the audit trail
Every event is SHA-256 chained to the previous one. The verify command exits clean only if the chain is intact — this is the evidence your customers ask for.
GET /v1/connections/{id}/audit
GET /v1/audit/verify?connection_id={id}
# → {"verified":true,"genesis_hash":"…","broken_at":null}Why not just a tool router?
Outbound tool routers meter API calls. Hebrah governs private data access — with evidence you can hand to your customers.
Inbound, not outbound
Tool routers point agents at SaaS APIs. Hebrah connects agents to the private data your customers already own — their Postgres, SFTP, FHIR, Notion, Slack.
Evidence, not just metering
A metered billing record says what an agent spent. A hash-chained audit export says exactly what it saw, scoped to the row — verifiable by your customer, not just by us.
Scoped and revocable, per connection
One global balance is convenient. A scoped, TTL-bound, instantly revocable per connection is safe to put in front of real data.
Outbound-only by design
Production Postgres joins through one outbound-only relay tunnel — no inbound firewall holes, and revoking a relay_id cuts access instantly. See the relay guide.
What your agent gets
20 targets, 6 packs
12 demo targets (audit GL ledgers, fintech settlements, synthetic FHIR, factory telemetry, workspace tools, developer sandboxes) plus 8 live account connectors — your own Stripe, Postgres, Notion, Slack, and more.
Scoped + TTL + revoke
Table-level and period-level scopes, TTL-bound connections, instant revocation. Agents never receive the underlying credential.
Hash-chained audit
Connection open and every query append to a SHA-256 chain. Export as JSONL, verify with one call.
$1 trial, no card
100 queries, 5 MB egress, 7 days. Then a prepaid wallet — $0.005 per query, $0.08 per GB egress. Pro at $99/mo when you scale.
Hosted MCP
One endpoint, every MCP host. The agent-side tools: create_account, discover_data_sources, connect_to_data_source, query_data_source, get_data_source_audit, get_connect_usage, revoke_data_source_connection.
Cursor
One-click Install in Cursor deeplink, or copy ~/.cursor/mcp.json config.
Claude
Claude Desktop and Claude Code: copy CLI or manual config with Bearer PAT.
Codex
OpenAI Codex CLI loads the hosted Hebrah MCP server.
OpenCode
Configure the hosted Hebrah MCP URL in OpenCode settings.
Windsurf
Copy serverUrl JSON for mcp_config.json.
Google Antigravity
Copy serverUrl JSON for mcp_config.json.
Zed
Zed MCP extensions with Bearer PAT auth.
VS Code
One-click vscode.dev MCP install redirect with Bearer PAT.
Any MCP host
Streamable HTTP hosted MCP with no local install required.
Start with $1 in free credit.
Agent loop: sign up headless in 90 seconds, discover targets, open a scoped connection, query, and export the hash-chained audit as your evidence trail. Give your agent /SKILL.md to onboard itself, or start in the dashboard at localhost:3000.