On this page

Find the shortest supported path

Find an app in the directory, connect an assistant, or build with Center's API. An API lets software request data or actions from a service. Assistants use Center's tools through MCP (Model Context Protocol); other integrations can use its web request interface, REST.

For a transaction, a plan stores the exact calls for review. Relay means submitting transactions the wallet has signed. Choose only the setup the task needs: browsing and MCP do not require an API account or bot. Keep the user's V6 project, network, wallet, asset, and amount explicit. See the glossary as technical terms arise.

What is ready

This review records the September 7, 2026 release state. Implemented means the documented path exists and has local checks; it does not mean a live wallet has completed it. Configured means the production service enables it. Live-checked identifies the narrower observations made against production. Always read capabilities for current availability before preparing an action.

  • Directory, API discovery, authenticated-route boundaries, and the 57-tool MCP catalog were live-checked. This checks availability, not every tool invocation or economic outcome.
  • Sponsored Safe-owner execution is configured on Ethereum (1), Optimism (10), Base (8453), and Arbitrum (42161). Provider quotes, gas estimates, sponsor signatures, and the deployed sponsor contract's identity were live-checked.
  • A real owner-signed sponsored V6 transaction and its verified outcome on each of those chains remain pending. Unsigned probes do not establish completion.
  • Recurring bot execution is implemented but inactive. It needs the matching deployed guard and owner-approved activation. It is optional for production owner execution. Operations records the checks.

Journeys and approvals

User goal Shortest supported path Necessary approval or prerequisite Status and completion boundary
Find an app, fund a project, or cash out Directory → Use Juicebox → chosen app. Use a known app link directly. No Center account. The app obtains wallet approval for its transaction. Directory live-checked. The destination app owns execution and version selection.
Learn, inspect code, or audit Directory → Audit or research → V6 source, deployed addresses, permissions, and documented guarantees. Assistants can use MCP source tools. No wallet approval. Verify the deployed code before applying what the source says. References implemented; source alone is not an audit or proof of what is deployed.
Launch a project, NFT project, or revnet Use the matching app builder, or its MCP launch planner with complete terms and a link to the project description. Review the economics and each wallet transaction. Publishing the description is a separate action. Supported core, 721, and revnet preparations; the chosen wallet or app handles execution.
Manage a project, shop, or revnet position Read current settings and permissions → prepare the change → approve it in the wallet → verify the result. Fresh approval for the exact action. Required earlier steps must be confirmed. Permissions belong to the wallet that acts. MCP journeys lists supported settings, payouts, routing, NFT tiers, token issuance, loans, and repayments; REST also exposes cataloged V6 contract methods.
Explore unfinished tools Directory → Explore unfinished tools → source or clearly labeled prototype. No production setup implied. eth.shop, Sticky, JBChat, JBProcessor, and Plugin are unfinished; they are excluded from the V6 protocol REST inventory.
Discover the REST interface API explorer → operation/schema and source → quickstart only when protected access is needed. None for OpenAPI, capabilities, or catalogs. Public discovery live-checked. A catalog entry alone does not establish a deployed target.
Automate protected reads or plan preparation Accounts → Sign in with email, phone, social account, or an existing wallet → API access → approve and download a connection → bot signs requests. The owner approves account enrollment and the exact bot grant. The bot's API permissions (its grant) have enforced scopes and expiry. Implemented. Bot reads and plans need no recurring wallet prompts; a grant gives no onchain spending authority.
Replace or revoke a bot Register a replacement key if needed → revoke the old grant on /accounts. Keep old plan IDs to check pending results. Fresh owner-signed grant change. Never upload a private key. Implemented. A replacement grant cannot take over the old grant's execution plans. Already admitted work may finish; revocation does not undo onchain authority.
Read a project now GET /projects/{chainId}/{projectId}?source=onchain, or /protocol/read for an exact method. Signed REST request; a registered read bot can sign automatically. Implemented with evidence from a block in the accepted chain (canonical). Failed reads remain unknown.
Search projects, activity, or history Use Bendystraw, an indexer that organizes chain records: project source=bendystraw, or cataloged /indexer/{entity} with an explicit network. Signed REST request. No personal Bendystraw key is needed through Center. Implemented. Follow page cursors; indexed values and USD estimates are not transaction quotes.
Execute from an ordinary wallet Prepare and simulate a REST plan → wallet signs each exact transaction → relay → refresh plan. Fresh owner-signed submission request, or bot request plus exact owner approval; wallet transaction signature is also required. Implemented. These transaction signatures lack timestamps, so fresh submission approval remains necessary. No smart wallet is required.
Execute with sponsored gas Create or reuse a supported Safe7579 wallet with one owner or a multisig threshold → bind it once → prepare its action plan and UserOperation → sign returned SafeOp → submit once → reconcile. Current Safe-owner threshold signs the validity-bound operation. API request authentication is separate. Creating a wallet currently costs owner gas. Four chains configured and provider probes live-checked; real signed V6 execution remains pending. No session guard or bot required.
Keep transaction funds separate or share approvals Optionally create or link a supported Safe7579 transaction wallet on Accounts → review its owners and approval threshold → fund it with a separately approved transaction. One owner can approve alone; a multisig requires its configured owner threshold. Linking grants no spending authority. Optional. Center verifies the supported wallet setup; arbitrary Safe configurations are not accepted.
Set up and transact on several networks Select wallet networks → review the shared setup → create and connect each wallet → add prepared actions to one transaction review list → follow each network's result. Each network's creation and transaction approvals remain explicit. One API account and bot connection work across the selected networks. Implemented. Matching wallet addresses have separate balances and permissions. Public recovery references survive reload; an unknown submission is checked before attempting another.
Execute several steps Keep one durable plan. For direct relay, confirm prerequisites and resume remaining steps. A Safe can batch eligible calls on one chain. Exact owner approval for each direct step or the selected Safe batch; never approve unknown future calls. Implemented. Multi-call Safe invocation can be verified, but modeled per-call economics currently remain unknown. Use one operation per modeled step when downstream work requires verified outcomes.
Use prepaid cross-chain relay Only when capabilities enable prepaid: plan → prepare wave → sign exact forward requests → publish → review/fund quote once → reconcile each chain. Fresh owner authorization, each forward signature, and a separately approved funding transaction. Implemented; availability is capability-dependent. Multiple calls per chain are supported within Center’s 32-step plan capacity. This is prepaid relay, not free gas or atomic settlement.
Follow a project across chains Start with one known V6 project → /projects/{chainId}/{projectId}/omnichain with the chosen source → inspect each chain's evidence. A bot-signed read. Claiming funds sent across a bridge needs its proof and a separate wallet-approved transaction. Implemented. Chain IDs/project IDs may differ; partial reads stay visible. Source confirmation, destination claim, credit to the project's payment contract, and final settlement remain distinct.
Let a bot repeat payments from a budget Optional: fund an isolated smart wallet → approve the exact bounded onchain permission → bot executes within it. Fresh owner approval of funding, finite allowances, policy installation, and changes. Inactive pending guard deployment and activation. An allocation record does not set aside or escrow actual funds. Seven/thirty-day durations are implemented choices, not API requirements.
Ask an AI agent to research or prepare Connect to https://juicebox.center/mcp → relevant tool → inspect the result. Add portable Juicebox skills when useful. No REST enrollment. Wallet actions and public metadata publication require their respective approvals. Transport and 57-tool catalog live-checked. MCP plans are unsigned and its tokens are not REST plan IDs. Any compatible MCP agent can connect.
Publish a project description (metadata) MCP jb_prepare_project_metadata → review exact JSON → jb_pin_project_metadata → use the returned link in the launch planner. Explicit approval to publish. No wallet needed for publication. Implemented for standard new metadata. A logo file is pinned first with jb_pin_project_logo and referenced as ipfs://<cid>; HTTPS logos are rejected because the webclients do not render them. Publishing a file does not launch or update a project.
Read or publish files through IPFS IPFS shares files by content; a CID identifies the content. Read with public GET /ipfs/{cid}; approved apps publish with /v1/pins/json, /file, or /media. No REST account for reads. Writes retain approved-origin and quota requirements; content becomes public. Existing shared services. Queuing another stored copy (pin) does not prove it is available. Limits.
Read raw chain data RPC is the request interface to a blockchain node. Send one read-only JSON-RPC request to public POST /v1/rpc/{chainId}. No REST account; read-method and request budgets apply. Existing shared gateway. It does not sign or broadcast transactions. RPC reference.
Build a client or add protocol coverage Choose REST schemas or V6 SDK actions → use current source examples from Juicebox Money, Revnet Money, or Juicescan → implement and test the feature. Wallet approval only for resulting execution; no approval needed to inspect reference code. Source-backed MCP integration plans and REST method catalogs implemented. Experimental apps and unsupported semantic adapters remain explicit.

Keep the path short without losing evidence

  • Ask for onchain versus bendystraw only where both serve the requested data. A client can remember the user's preference. Do not query both unless comparing them or resolving a real discrepancy. Preparation always checks onchain state.
  • Reuse an existing account, selected source, wallet link, or plan when still valid. Do not repeat enrollment, create a new smart wallet, or require sessions for an ordinary transaction. Keep the same API owner or bot through submission.
  • Show one concrete review of the exact calls, amounts, recipients, fees, and validity before the necessary signatures. API authentication, transaction authority, and publication approval protect different actions.
  • Check an uncertain submission against chain evidence before offering another transaction. Keep its existing hash, operation ID, and idempotency key. A pending result is not an instruction to pay again, change transport, or repeat publication.
  • A directory decision must reach a usable resource. Return links are optional after a resource; questions cannot loop indefinitely. The shared graph validator and navigation checks enforce this across views.