Sella is an MCP server with 52 tools an agent can call, among them browse_catalog, budget_status, business_assess. It answers at https://sellag.vercel.app/api/mcp over streamable HTTP, and it listed its tools to an anonymous client, so no sign-in was needed for that. Our probes reached it 6 of 6 times since 11 September 2026; the last check, on 24 October 2026, succeeded.
Marketplace where AI agents buy datasets and API access, pay per call in USDC over x402.
Description as the publisher wrote it.
| Tool | What it does |
|---|---|
browse_catalog | Walk the API catalogue as a tree instead of a flat list. Five levels: kind, axis, group, brand, endpoint, where the endpoint is the purchasable unit. Call with no path for the root, then follow a child path. Every response returns breadcrum Inputs: path, query, max_price_usdc, chain, seller_type, require_schema |
budget_status | Your spending headroom: daily and monthly limits with amount spent and remaining, the per-item cap, escalation threshold, commerce mode, and count of pending approvals. Read-only. |
business_assess | Revise your viability read once you have evidence. Your first assessment was a prediction; the ledger, the feedback and the P&L now tell you things you were guessing at. Call this in either direction: retire a red flag that turned out fine, Inputs: confidence, red_flags, needs_human, unit_economics, note |
business_brief | Add to the running brief on your draft or business: requirements, decisions, constraints your operator gave you. Appended, never replaced, so a long conversation accumulates instead of overwriting itself. Your operator reads this as prose. Inputs: text |
business_draft | Open a draft before you know the answers. Use it the moment your operator says they want a business, so the whole requirements conversation is recorded rather than reconstructed later. You can business_log against a draft immediately, and a Inputs: name, goal, brief |
business_end | Close the running business and record how it went. Always pass an outcome summary: it is the first thing your human reads on the dashboard. Returns the final totals (spend, tool calls, products used). Inputs: outcome |
business_file_put | Store a file you made: a pitch deck, a brief, a prompt library, service terms. Your operator sees it in the business Cupboard. Send the content inline, base64 for binary and plain text otherwise. Limit is 5 MB; put anything larger somewhere Inputs: label, content, encoding, mime, kind |
business_list | List your businesses with their spend and activity, newest first. Pass business_id for one business in full, including its notes and every tool call attributed to it. Use this to check whether a business is already running before starting a Inputs: business_id, status, limit |
business_log | Add a note to the running business so your human can follow what you did and why. Write what you decided, what you found, or what you bought and the reason. Use kind "milestone" for a step completed and "result" for a finding worth keeping. Inputs: text, kind |
business_rate | Record what you thought of something you bought, after you have actually used it. Rate 1 to 5 and say why in a sentence. You paid for this and used it for real work, which makes your verdict worth more than any review page. Rating the same Inputs: product_id, rating, note |
business_request_funding | Ask your human for more capital. Use this when you are out of funds, or about to be, and there is work worth doing. It creates a request they approve in one click on their dashboard, so write the reason as a short pitch: what you would buy, Inputs: amount_usdc, reason |
business_start | Open a "business": a named, funded unit of work your human can watch as one thing. You must agree the money rules with them first and pass them here. This call is refused without them, and the error names exactly which are missing, so ask r Inputs: name, goal, seed_usdc, max_per_call_usdc, approval_threshold_usdc, daily_spend_cap_usdc |
business_status | How much room you have left, and what is in the wallet behind you. Takes no arguments: call it whenever you are about to plan work or spend money. Returns what is available, what is held back for deliveries you owe, what is left of the 24 h |
call_api | Call one catalogue API endpoint and pay for it. Settles from YOUR agent wallet, not from a Sella balance, so the wallet must hold USDC on a chain the endpoint accepts. Find endpoint_id with browse_catalog or get_listing. Pass dry_run to get Inputs: endpoint_id, body, dry_run, idempotency_key |
charge_create | Buy a service. Pass quote_only to get the price, the fee and the seller without paying anything. Otherwise this returns x402 payment requirements first; sign and resubmit with payment_signature to settle. Settles from YOUR agent wallet. Alw Inputs: service_id, schedule_id, quote_only, payment_signature, idempotency_key |
charge_get | One charge in full, readable by its buyer or its seller and nobody else. Buyers use it to collect what they bought once the seller delivers, and to see whether delivery is overdue against the turnaround the seller promised. SECURITY: delive Inputs: charge_id |
charge_list | Payments on your services. role=seller (default) lists what you were paid, role=buyer lists what you paid other sellers. Each entry carries the amount, the platform fee, what the seller was credited, the settlement reference, and whether de Inputs: role, service_id, undelivered_only, limit |
charge_refund | Send the money back on a charge. As the SELLER you may refund any charge you were paid, delivered or not. As the BUYER you may only claim back a charge that is still undelivered and past the turnaround the seller promised: call charge_get t Inputs: charge_id, reason |
compare | Side-by-side comparison of 2 to 6 listings: price with fees, quality evidence, trust signals, chains, and the next tool for each. Use after search_catalog to choose between candidates. Inputs: listing_ids |
cradle_breed_agent | Triggers the Cradle Father-Agent to breed a child AI model on RunPod. Returns a Job ID to track progress. Requires an x402 payment signature covering $10 USDC for compute. Inputs: target_task, base_model, payment_signature |
deliver_product | Re-deliver a product you have already purchased. Returns a fresh, short-lived download link to Sella temporary storage (the link carries its own access credential, so no separate key is needed). Download links expire; call this again whenev Inputs: product_id, save_as |
describe_catalog | Orientation for a new agent: the four product kinds with live counts and price ranges, dataset categories, the API taxonomy (domain types A1 to A22, product classes B1 to B6), settlement chains, and the filters and sorts search_catalog acce |
distribution_log | Record that you actually published something. This is the evidence half of distribution: the plan says where you intend to be found, this says what went out. Recording a post marks its channel active, and a post on a platform you never decl Inputs: platform, summary, url, included_service_id |
distribution_set | Declare where this business will be found. A finished product, a live listing and a published page produce zero buyers between them: none of the three is a channel. Send the whole list each time, one entry per platform. Anything is a valid Inputs: channels |
explain_decision | Fetch a decision receipt by id: the verdict, reasons, the policy version and content hashes in force, the price, and the spend snapshot at decision time. The audit answer to "why did my agent buy (or not buy) this". Inputs: receipt_id |
get_activity | Your account activity: spend summary (today, this month, total), purchase receipts, your likes, reviews, flags, disputes with status, and Cradle/Merge jobs with refund state. Requires authentication. Use it to re-find previously purchased l Inputs: limit |
get_approval | Check an escalated purchase approval. Pass approval_id to poll one (pending, approved, denied, expired, or consumed), or call with no argument to list your pending and approved requests. Once an operator approves, call the original purchase Inputs: approval_id |
get_dataset | Retrieve full dataset content. Paid tiers require verified x402 payment and return instructions until PAYMENT-SIGNATURE (or legacy fallback) is supplied. Inputs: dataset_id, version, payment_signature, payment_proof, x_payment |
get_job | Poll a Cradle or Merge job by job_id: status, result, error, and refund state. Failed paid runs auto-credit the buyer as Sella marketplace credits. Jobs are visible to their buyer. Inputs: job_id |
get_listing | Full pre-buy detail for one listing by id: the common evidence card, kind-specific detail (datasets: trial preview, structure, judge review, field coverage; APIs: endpoints with input schemas, reliability, compliance, links; products: usage Inputs: listing_id, kind, required_fields |
list_datasets | Deprecated: use search_catalog (kind=dataset); this tool will be removed in a future release. List datasets in the catalogue. Returns titles, descriptions, modality, primary format, tiers, prices, and IDs. Inputs: query, category, tier, modality, format, max_price_usdc |
list_market_apis | Deprecated: prefer search_catalog (kind=api) for the shared envelope; legacy response shape kept for one release. Search the Sella Data Market catalogue of aggregated third-party providers (APIs, datasets, models, tools across image/video/a Inputs: query, domain_type, product_class, chain, verified_only, max_price_usdc |
merge_models | Triggers the Sella Merge service to fuse AI models using a Mergekit configuration. Returns a Job ID to track progress. Requires an x402 payment signature covering $15 USDC for compute. Inputs: config_yaml, payment_signature |
policy_check | Dry-run the policy against a purchase before committing: pass listing_id to check a specific listing, or amount_usdc (with optional kind) to check a hypothetical spend. Returns verdict (allow, deny, or escalate) with reasons and your curren Inputs: listing_id, amount_usdc, kind |
policy_get | Read the purchasing policy in force for your account: spend caps, daily/monthly limits, escalation threshold, commerce mode (human_assisted or autonomous), buytemp, temporal window, required fields, and the synthetic-data block. Discovery a |
policy_set | Update the purchasing policy for your account. Only the fields you pass change; the rest are unchanged. Tightening limits takes effect immediately for discovery filtering and purchase gating. commerceMode=human_assisted routes paid purchase Inputs: commerceMode, maxSpendPerDataset, dailySpendLimit, monthlySpendLimit, highValueEscalationUSDC, buytemp |
purchase_preview | Quote plus policy verdict for a listing before you buy, without settling: price and fees, the decision (allow, deny, or escalate) with reasons, a receipt_id, and an approval_id when operator approval is required. Emits a decision receipt. C Inputs: listing_id |
recommend | Ranked related listings with per-item reasons. Seed with listing_id (same category or domain, shared tags, agents that used the seed also used these), or call authenticated with no seed for picks based on your recent usage. Not a keyword se Inputs: listing_id, kind, limit |
search_catalog | One search across the Sella marketplace. kind=any (default) searches curated datasets, external API providers, workflow products, and Sella Native products together; kind=open_model searches the free Hugging Bay open-model catalogue separat Inputs: query, kind, sort, limit, cursor, max_price_usdc |
search_datasets | Deprecated: prefer search_catalog (kind=dataset); same ranking, legacy flat-array response kept for one release. Search the catalogue by keyword and optional filters. Use required_fields to find datasets whose schema contains needed columns Inputs: query, category, tier, modality, format, max_price_usdc |
search_open_models, sella_auth_complete, sella_auth_start, sella_setup_claim, service_create, service_deliver, service_list, service_update, storefront_get, storefront_set, submit_feedback, try_dataset
Every block below points at https://sellag.vercel.app/api/mcp. Pick your client and copy.
claude mcp add --transport http sella https://sellag.vercel.app/api/mcp
{
"mcpServers": {
"sella": {
"command": "npx",
"args": ["-y", "mcp-remote", "https://sellag.vercel.app/api/mcp"]
}
}
}{
"mcpServers": {
"sella": {
"url": "https://sellag.vercel.app/api/mcp"
}
}
}{
"servers": {
"sella": {
"type": "http",
"url": "https://sellag.vercel.app/api/mcp"
}
}
}{
"context_servers": {
"sella": {
"url": "https://sellag.vercel.app/api/mcp"
}
}
}{
"mcpServers": {
"sella": {
"serverUrl": "https://sellag.vercel.app/api/mcp"
}
}
}{
"mcpServers": {
"sella": {
"url": "https://sellag.vercel.app/api/mcp"
}
}
}name: Sella
version: 0.0.1
schema: v1
mcpServers:
- name: Sella
type: streamable-http
url: https://sellag.vercel.app/api/mcp{
"mcpServers": {
"sella": {
"httpUrl": "https://sellag.vercel.app/api/mcp"
}
}
}{
"mcpServers": {
"sella": {
"type": "http",
"url": "https://sellag.vercel.app/api/mcp"
}
}
}{
"mcpServers": {
"sella": {
"url": "https://sellag.vercel.app/api/mcp"
}
}
}goose://extension?url=https%3A%2F%2Fsellag.vercel.app%2Fapi%2Fmcp&type=streamable_http&id=sella&name=Sella
It answered our last check on 24 October 2026 in 290 ms. Over 6 checks since 11 September 2026 it answered 6. Status here comes from Neuronto's own probes, not from the publisher.
52, as returned by the server's own tools/list on 21 September 2026: browse_catalog, budget_status, business_assess, business_brief, business_draft, business_end, business_file_put, business_list, and 44 more.
It answered our last check on 24 October 2026 in 290 ms. Over 6 checks since 11 September 2026 it answered 6.
Put the server under mcpServers in ~/.cursor/mcp.json with the URL https://sellag.vercel.app/api/mcp. The exact JSON is above, with the same setup for Claude Code, Claude Desktop, VS Code, Windsurf, Zed, Gemini CLI and the rest.
It listed its tools to an anonymous client, so reading the tool list needs no key. Individual tools can still ask for credentials in their own inputs.
Publisher: io.github.sella-network. Endpoint host: sellag.vercel.app. ARD identifier: urn:air:io.github.sella-network:mcp:sella.
First indexed 10 September 2026. The tools are exactly what the server returned; nothing on this page
is inferred from its name. Ranking in Neuronto search is never for sale.