origo-bc-accounting

Origo BC — MCP operating rules

Follow these rules whenever the user is working with the Origo BC MCP endpoint (https://dynamics.is/api/mcp), skills, prompts, UBL templates or the /origo-bc-* commands.

Full accounting rules — get_bc_accounting_rules

This skill is a compact starter set. The MCP server hosts a comprehensive version of the Origo BC accounting and development rules at https://origopublic.blob.core.windows.net/resources/mcp/CLAUDE.online.md.

When this skill loads, call get_bc_accounting_rules on the MCP server to fetch the full rules. The tool supports three retrieval modes:

Call	What you get
get_bc_accounting_rules()	Frontmatter, intro, and a heading-only table of contents (small payload — start here).
get_bc_accounting_rules({ section: "AL Naming Conventions" })	A single section by heading text (case-insensitive, substring match).
get_bc_accounting_rules({ full: true })	The entire document — use only when the full context is truly needed.

The tool caches the document for 5 minutes. Pass { refresh: true } to bypass the cache.

Recommended flow:

1. Call with no arguments to get the TOC.
2. Load sections on demand as the conversation requires them.
3. Only request { full: true } when a broad review or cross-cutting task needs the complete document.

Cowork / Claude.ai scope

When running inside Cowork chat or Claude.ai (not VS Code / Claude Code):

• Use only get_config and set_config to read and write skills and prompts.
• Never call check_standards_status, update_bc_standards, or setup_origo_bc_environment — those tools are for VS Code's developer environment only.
• Skills and prompts live in the Business Central database, not in local files.
• The words "SKILL" / "skills" always refer to BC MCP-Skills records.
• The words "PROMPT" / "prompts" always refer to BC MCP-Prompts records.

Two BC namespaces

Source	Role	Content
MCP-Skills	Knowledge / lookup text	JSON with patterns, rules, reference data
MCP-Prompts	Triggers / workflows	Markdown body + frontmatter + variables

Both namespaces use the same index pattern: the record at GUID 00000000-0000-0000-0000-000000000000 is the index of every record in that namespace.

Fetch the index only when the user asks about BC, skills, or a BC-adjacent task — never automatically at the start of every conversation.

Reading skills from BC

get_config(source: "MCP-Skills", id: "00000000-0000-0000-0000-000000000000")
get_config(source: "MCP-Skills", id: "<guid>")

data is the full record body — use it as source material for the relevant slice of work.

Two skill storage types

type	Where is the content?	How to load
content	Inline in the record's data field	Read data directly
remote-loader	Metadata + sourceUrl only; body is on the web	Fetch sourceUrl via WebFetch and treat the response as source

For a remote-loader skill: fetch the record, then fetch sourceUrl, then use the downloaded content as the authoritative body.

Reading prompts from BC

Prompts are slash-command templates that reference one or more skills via skillRefs.

get_config(source: "MCP-Prompts", id: "00000000-0000-0000-0000-000000000000")
get_config(source: "MCP-Prompts", id: "<guid>")

Prompt record shape:

{
  "name": "<prompt-name>",
  "description": "<what it does>",
  "frontmatter": { "mode": "agent", "tools": [] },
  "variables": [ { "name": "foo", "type": "input", "description": "" } ],
  "skillRefs": ["<skill-guid-1>", "<skill-guid-2>"],
  "body": "<markdown with ${input:variable} placeholders>"
}

When a user's request matches a prompt: fetch that prompt, fetch every referenced skill, then execute the body.

MCP connection formats — two options

The server's resolveConn accepts two formats for encryptedConn (whether the blob arrives as the x-encrypted-conn header or as a tool argument):

Option A — plain:<base64> (preferred for new installs)

Base64-encoded JSON {tenantId, clientId, clientSecret, environment}. No HTTPS setup round-trip, no MCP_ENCRYPTION_KEY required server-side.

How to build the blob:

• Windows: .\Create-PlainConnectionString.ps1 -TenantId ... -ClientId ... -Environment ... → returns dpapi:<base64> (the plain: payload DPAPI-wrapped to the current Windows user/machine).
• macOS / Linux: node create-connection-string.js --tenant ... --client ... --environment ... → returns plain:<base64> with no DPAPI wrap; protect the config file with chmod 600 or a platform secret manager.

Protection in practice:

• In transit: TLS from proxy to server.
• At rest: DPAPI on Windows (bound to user + machine), filesystem permissions elsewhere.
• No double encryption, no server key to leak.

Option B — AES-256-GCM ciphertext (legacy, still supported)

Calls encrypt_data on the server with plaintext JSON → receives base64 ciphertext → DPAPI-wraps locally on Windows.

Use only for environments that already have configured dpapi:<...> blobs containing AES ciphertext, or when the user explicitly requests a second encryption layer.

Choosing

Situation	Choice
New install on Windows	A (plain: + DPAPI)
New install on macOS / Linux	A (plain: + chmod)
Existing working config	B — leave it alone
Server without encryption key	A

The stdio-proxy.js bridge is format-agnostic — it forwards whatever string it has in x-encrypted-conn. The server branches on the prefix (plain: decodes base64, otherwise AES-decrypts).

Update rules

When you learn something new

If you discover a new pattern, a known bug, a workaround, or a confirmation that something works (or doesn't), write it into the relevant skill record immediately so it persists across sessions.

set_config(source: "MCP-Skills", id: "<guid>", data: { ...updated body... })

When adding a new skill

1. Mint a new GUID in the form AXXXXXXX-0000-0000-0000-XXXXXXXXXXXX:
   • A1xxxxxx — bc-general/* (content)
   • A2xxxxxx — bc-journal-corrections/* (content)
   • A3xxxxxx — bc-incoming-document/* (content)
   • A9xxxxxx — top-level orchestration skills (content)
2. Decide type: content (store in BC) or remote-loader (store metadata + URL).
3. Save the skill with set_config(source: "MCP-Skills", ...).
4. Update the skills index with the name, GUID, type, description, and trigger conditions.

set_config(source: "MCP-Skills", id: "00000000-0000-0000-0000-000000000000", data: { ...index with new entry... })

When adding a new prompt

1. Mint a new GUID in the form BXXXXXXX-0000-0000-0000-XXXXXXXXXXXX:
   • B9xxxxxx — top-level orchestration prompts (pairs with A9 skills).
   • GUIDs may contain only hex (0-9, A-F). P is not valid hex.
2. Save the prompt with set_config(source: "MCP-Prompts", ...).
3. Update the prompts index with the name, GUID, description, variables, skillRefs, and trigger conditions.

When to update the index

Always update the index when:

• A new record is added.
• A record is removed or merged into another.
• A trigger condition or description changes.

UBL XML templates

Icelandic PEPPOL UBL XML templates live in Cloud Events Storage under source = "UBL Templates". Whenever Claude is producing PEPPOL UBL XML (invoice, credit note, order, etc.), fetch a template first. Never write UBL XML from scratch; the templates enforce correct namespace declarations, PEPPOL CustomizationID / ProfileID, and Iceland-specific defaults.

get_config(source: "UBL Templates", id: "DDDD0000-0000-0000-0000-000000000000")
get_config(source: "UBL Templates", id: "<guid-from-index>")

Three standards are supported:

Standard	UBL version	Use
PEPPOL BIS 3.0	2.1	Current standard — use by default
PEPPOL BIS 2.0	2.1	Transition standard (~2017–2021)
IS e-reikningur BII1	2.0	Original Icelandic electronic invoice (legacy systems)

Identify the right standard from CustomizationID and ProfileID on the source document.

Template use:

1. Fetch the template for the document type.
2. Replace every {{placeholder|default}} with the real value (use default if nothing is available).
3. Remove empty optional blocks (those whose placeholder had no default and no data).
4. Repeat line blocks (InvoiceLine, CreditNoteLine, etc.) once per line.
5. Resolve customer data by field numbers 90 (GLN) and 47 (Registration No.) to determine EndpointID.

EndpointID resolution

GLN value	EndpointID	schemeID (BIS 3.0)	schemeID (BIS 2.0)	schemeID (BII1)
10 digits	GLN (= kennitala)	0196	9917	IS:KT
13 characters	GLN (international GS1)	0088	0088	0088
Empty	Registration No. (kennitala)	0196	9917	IS:KT

0088 = GS1 GLN scheme. 0196 = Icelandic national registry (BIS 3.0). 9917 = kennitala EAS code (BIS 2.0). IS:KT = kennitala (BII1).

Local files

Files under C:\Data\MCP\prompts\*.prompt.md are mirrors of BC prompt records — VS Code / Claude Code reads them as slash commands. They are not the source of truth; if a mirror drifts from BC, rebuild it from the BC record.

Single source of truth

Skills and prompts in BC are the one source of truth. Local files only describe how to fetch them — they don't store the content.