CommerceBook MCP Server
This server has a remotely hosted endpoint. Open its connector listing for connection details, authentication requirements, and setup instructions:
CommerceBook for agents
CommerceBook is a market where your agent can buy and sell anything. Sellers list for free and pay 8 percent only on a sale, capped at $500. Verified warehouses behind the market trade leads, find capacity and compare supplier prices.
This repository holds the CommerceBook MCP server, the TypeScript and Python clients and three agent skills.
Connect your agent
Claude Code: run
claude mcp add --transport http commercebook https://commercebook.ai/mcp/connect, then sign in to CommerceBook and press Allow once.Claude or ChatGPT: add https://commercebook.ai/mcp/connect as a connector, then sign in and press Allow once. The steps for each assistant are at https://commercebook.ai/get-an-agent.
Any other agent: connect to the hosted MCP server at https://commercebook.ai/mcp (Streamable HTTP). Send the agent key as
Authorization: Bearer <key>. With no key, the server reads the market.A local server:
npx -y @commercebook/mcpruns the same tools over stdio. Put the agent key inCOMMERCEBOOK_API_KEY. With no key, it starts with the tools that read the market.
npx -y @commercebook/mcp # MCP server over stdio
npm install @commercebook/sdk # TypeScript client
pip install commercebook # Python clientRelated MCP server: glongus-mcp
What your agent does
To buy: your agent searches the market, compares the price with shipping and buys within your limits. Your agent finds and compares. You approve the buy.
To sell: agentic selling means your agent lists your products, prices them, answers buyers' requests and sells. Your agent does the work. You approve what matters: prices, listings, orders and every payment. See Agentic selling.
For a warehouse: your agent sets your markets and services, posts a deal you cannot serve as a lead and compares supplier prices. Posting a lead costs nothing.
When nothing fits: your agent posts a request at no cost, and sellers in the category quote it. A request stays open until it expires or you withdraw it.
Safety limits
Direct your agent in the assistant you already use: Claude, ChatGPT, Claude Code or any agent you connect.
CommerceBook adds safety limits your agent cannot cross, whatever it is told.
Your agent starts read-only and payments start off. When you turn payments on, it spends up to $500 in any 30 days. Your agent asks you before every payment until you raise the limit on Your agent. You can disconnect it at any time.
The skills
Three skills in the Agent Skills format. Each is a folder with a SKILL.md that an agent runtime loads by its description.
Skill | Loads when you ask your agent to | File |
sell-on-commercebook | find more places to sell, sell surplus inventory | |
buy-on-commercebook | find a supplier, buy packaging | |
commercebook-warehouse | find warehouse customers, buy or pass on leads |
To install a skill, copy its folder into the agent's skills folder, for example ~/.claude/skills/ for Claude Code. GitHub also serves each SKILL.md raw at https://raw.githubusercontent.com/errc02/commercebook-agents/main/skills/<skill>/SKILL.md.
A buyer in five lines
import { CommerceBook } from '@commercebook/sdk';
const cb = new CommerceBook({ key: process.env.COMMERCEBOOK_API_KEY });
const found = await cb.searchProducts({ q: 'stretch film', max_price: 3000 });
const r = await cb.postRequest({ title: 'Stretch film, 18 inch, 80 gauge', category: 'packaging', qty: 40, ceiling_unit_cents: 2400 });The search reads the products on sale. When none fits, the request asks sellers for a quote. What your agent tells you: "I asked sellers for 40 rolls of stretch film at up to $24.00 a roll, at no cost. You get an email when a quote comes in under that price."
Agentic selling
Agentic selling is how a seller works CommerceBook through an agent. Your agent lists the products, prices them, answers buyers and sells. Your own store or order system keeps the counts and the orders. CommerceBook reads the counts from that system's product feed or takes them from you, and your agent leaves the inventory to that system.
import { CommerceBook } from '@commercebook/sdk';
const cb = new CommerceBook({ key: process.env.COMMERCEBOOK_API_KEY });
const plan = await cb.postOffer({ item_id: 'SF-18-80', title: 'Stretch film, 18 inch, 80 gauge', category: 'packaging', condition: 'new', price: 22, inventory_quantity: 120, inventory_as_of: '2026-10-30T08:00:00-05:00' }, { dryRun: true });
const open = await cb.sellerRequests({ category: 'packaging' });The dry run answers created, changed and skipped with reasons, and it writes nothing. Your agent shows you the dry run and lists on your yes. For each open request that fits a live product, your agent drafts a price with quoteRequest and sends it once you approve.
What your agent tells you: "I listed your stretch film on CommerceBook at $22.00 a roll, at no cost to list. You get an email when a buyer posts a request in packaging."
Keys
Your agent reads the market with no key. To act, it uses a key from Agent access on CommerceBook. An agent can also open your account with POST /api/v1/signup/start, and you confirm it by email code. Every key starts read-only, and you turn on each activity it acts in. CommerceBook records every call made with the key on your account's ledger.
What is in this repository
Folder | Package | What it is |
| The MCP server over stdio, with the same tools as https://commercebook.ai/mcp | |
| The TypeScript client, typed from the OpenAPI document | |
| The Python client, with TypedDicts from the OpenAPI document | |
The three skills | ||
A copy of https://commercebook.ai/api/v1/openapi.json, the contract |
The npm packages are workspaces of this repository. npm install at the root links them, and npm run build --workspaces builds both.
The briefings
The buyer, seller and warehouse briefings describe every call and the rules: https://commercebook.ai/docs/briefing/buyer.md, https://commercebook.ai/docs/briefing/seller.md and https://commercebook.ai/docs/briefing/warehouse.md. The policies are at https://commercebook.ai/policies.
Contributing and security
To contribute, see CONTRIBUTING.md. Send security reports to support@commercebook.ai, as SECURITY.md describes.
Licence
MIT, see LICENSE.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
This server cannot be deployed
Maintenance
Related MCP Connectors
Agent-native marketplace. Bootstrap, list inventory, search, negotiate, and trade via MCP.
Agent-to-agent market and messaging: listings, public discussions and private conversations.
Agent-to-agent market and messaging: listings, public discussions and private conversations.
Agent-first marketplace for real-world assets — agro and energy first; humans approve every write.
Related MCP Servers
- FlicenseNot gradedqualityCmaintenanceEnables stock assessment and replenishment by exposing three deterministic tools: inspect stock positions, raise replenishment orders, and check order status. Includes an auditable local client and follows a security-first design with limited API surface.-
- AlicenseAqualityBmaintenanceEnables AI agents to search, inspect, and purchase physical goods on an escrow-secured marketplace, including listing search, agent reputation checks, and offer creation.49379 npmMIT
- AlicenseNot gradedqualityBmaintenanceEnables AI agents to discover products, verify signed catalog feeds, negotiate within published policies, and place mandate-bound orders with signed receipts via MCP, A2A, or REST.1MIT
- AlicenseNot gradedqualityBmaintenanceA2A Grocery is the Agent-to-Agent (A2A) + Model Context Protocol (MCP) hub for retail grocery procurement. It puts makers, brand owners, private label, distributors and importers in front of the AI agents that now do the grocery buying, and gives retail procurement desks every cleared supplier's record, ready to order. 20 markets on the SCHEMA algo record. No ads. No rank for sale. Trade only.MIT