Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    b1-mcp-server: SAP Business One context for AI agents

    1

    Connect AI agents to SAP Business One HANA/SQL databases and Service Layer with built-in read-only guardrails.

    Free

    1 useSecurity scanned
    b1-mcp-server: SAP Business One context for AI agents

    b1-mcp-server: SAP Business One context for AI agents

    Example session with this skill installed

    Create an inventory count for item ITEM-1001 in warehouse 01, counting the same quantity that is on hand, then close it.

    • Read your context and instructions
    • Compiled the b1-mcp-server

    Summary

    I created the inventory count through the Service Layer and closed it, with your approval for each write. The counted quantity matches the quantity on hand, so the count has no variance.

    What happened

    StepCallResult
    1. Check the connectionconnect_databaseServiceLayer write approval: client (the MCP client's permission prompt; no approval form)
    2. Read the stockSELECT OITW."ItemCode", OITW."WhsCode", OITW."OnHand" FROM OITW WHERE OITW."ItemCode" = 'ITEM-1001' AND OITW."WhsCode" = '01'On hand: 120
    3. Create the countPOST InventoryCountingsYou approved the prompt. Created DocEntry 6, DocNum 4
    4. Close the countPOST InventoryCountings(6)/CloseYou approved the prompt
    5. ConfirmGET InventoryCountings(6)?$select=DocumentStatuscdsClosed

    The body sent in step 3

    {
      "BranchID": 1,
      "CountDate": "2026-09-28",
      "Remarks": "Inventory count created by the agent",
      "InventoryCountingLines": [
        {
          "ItemCode": "ITEM-1001",
          "WarehouseCode": "01",
          "CountedQuantity": 120,
          "Counted": "tYES"
        }
      ]
    }
    

    Each write named its target database and showed the exact endpoint and body before you approved it. Nothing was written through SQL: the database side stays read-only.

    Next steps

    • Open the count in SAP Business One (Inventory › Inventory Transactions › Inventory Counting) to review it.
    • To record a real difference, send the counted quantity instead of the on-hand value; posting the difference is a separate document (InventoryPostings).
    • For many items, send one count with several lines instead of one write per item.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Querying SAP Business One directly through AI agents often hits a wall due to database engine differences and strict read-only requirements. This skill teaches your agent to use b1-mcp-server (github.com/bersimao/b1-mcp-server), an open-source MCP server that bridges AI agents and SAP B1 on HANA or MS SQL, plus the Service Layer, behind production-safe guardrails.

    What it does

    • Database Connection guides the agent through your connection profiles for HANA and MS SQL, including the Service Layer OData API, and stops it after failed logins instead of locking the account.
    • Read-Only SQL keeps the agent on SELECT queries and read-only blocks, and has it hand any data change back to you instead of trying to work around the server.
    • Schema Discovery points the agent to the server's metadata tool for tables, columns, and procedures, instead of hand-written catalog queries.
    • Service Layer Access covers OData GET, plus POST, PATCH, and DELETE when your profile allows them, with the server's approval workflow for each write.
    • Guardrail Handling explains every refusal, engine-specific timeout, query length limit, and truncated result, so the agent fixes the query instead of retrying blindly.

    How it works

    1. Use connect_database with "list" to view available B1 profiles and their connection capabilities.
    2. Select a profile ID to establish a connection to the database, Service Layer, or both.
    3. Run execute_sql for data retrieval or execute_service_layer_write for approved record updates.
    4. Watch tool output for TRUNCATED notes or engine-specific timeouts, and refine the query.

    Frameworks & tools

    Built for the Model Context Protocol (MCP) and b1-mcp-server. Compatible with SAP Business One Service Layer (OData v3/v4), SAP HANA, and Microsoft SQL Server.

    Why this beats prompting it yourself

    Generic prompts don't know that a HANA timeout shows up as Socket recv timeout while MS SQL reports a plain request timeout, or which refusals can be fixed and which must go to a human. This skill gives the agent exact error meanings, readable SAP B1 SQL conventions (full table names instead of T0/T1 aliases), and the write rules that keep ERP data safe.

    Use cases

    • Fetching real-time stock levels or Business Partner balances for analysis.
    • Auditing user-defined tables (UDT) and system tables like OINV or OCRD.
    • Updating documents via the Service Layer, with your approval for each write.
    • Debugging SAP B1 stored procedure logic by rewriting it as a SELECT.

    What you get

    One SKILL.md file. Copy it into your agent's skills folder, for example ~/.claude/skills/b1-mcp-server/ for Claude Code. Requires b1-mcp-server (npx b1-mcp-server). Open source (MIT); the same skill lives in the GitHub repository.

    Known limitations

    No PUT, no $batch, and no SQL writes. Queries are capped at 8000 characters, and results at 500 rows and 100,000 characters by default. Service Layer writes are off unless your profile enables them.

    Requires b1-mcp-server; the skill does nothing on its own. No SQL writes, DDL, or stored procedure calls (EXEC/CALL), and no OData PUT or $batch. Queries are capped at 8,000 characters, and results at 500 rows and 100,000 characters. On HANA, a query that exceeds the timeout (60 s by default) shows up as a dropped connection, not as a timeout. Service Layer writes are off until a verb is enabled in the profile, and by default each write needs approval through an MCP form, which some clients cannot show. One company connection at a time.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    1 use

    Times used to date

    Free forever

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 29 days ago

    • Free with an Agensi account

    Requires

    B1 MCP
    Node.js 20+
    SAP Business One

    Permissions

    Network Access

    The skill itself only gives instructions to the agent. All access goes through b1-mcp-server: read-only SQL, and Service Layer writes only when the profile allows them and the user approves each one.

    Listed29 days ago

    What's inside

    Frequently Asked Questions

    Each connection profile declares its engine (dbType "hana" or "mssql"), and the server uses the matching native driver. Schema discovery runs engine-specific catalog queries, and the read-only guardrails cover both dialects (CALL and EXEC, DO BEGIN and BEGIN blocks). Your agent still writes SQL in the engine's own syntax; the skill tells it what differs, such as HANA reporting a timeout as a dropped connection while MS SQL reports a plain request timeout.

    The database side is read-only, enforced by the server rather than by the prompt: only SELECT and read-only blocks run, and anything else is refused. On the Service Layer, profiles allow only GET by default; each write verb must be enabled per profile, and by default every write needs your approval. Reads still put load on the database, so queries are stopped after 60 seconds by default, and lock hints that could block users are refused. Keep production profiles on the default approval mode, and prefer a database user with read-only permissions.

    Any MCP client that runs local (stdio) servers. It is developed and tested with Claude Code. Service Layer writes in the default mode need a client that supports MCP approval forms (elicitation); without one, writes are refused. The skill itself is a standard SKILL.md file for agents that read skills.

    Yes, for reading. User-defined fields (U_ columns) and UDTs ("@" tables) can be queried with SELECT like any other table, and schema discovery lists them. Through the Service Layer, user-defined fields travel in entity bodies, and UDOs are reached through their own endpoints, with writes subject to the same verb and approval rules.

    Node.js 20 or newer, the server added to your MCP client (npx -y b1-mcp-server), and a connections file at ~/.claude/connections.json (or the path in MCP_CONNECTIONS_FILE) with one profile per company: database host, name, user and password, and/or the Service Layer URL, user and password. The machine running the agent needs network access to the database and/or the Service Layer. Credentials stay in that file; the server never sends them to the agent.

    Yes, all adjustable with environment variables. Queries are limited to 8,000 characters, results to 500 rows and then 100,000 characters (always announced with a TRUNCATED note), and each query to 60 seconds. Each tool allows 60 calls per minute. Service Layer write bodies are limited to 50,000 characters.

    Not through SQL: the server is read-only. Service Layer writes only work if your profile allows the verb, and by default each one needs your approval.

    No. It teaches the agent how to use b1-mcp-server's tools.

    New in LLM Ops