Skip to main content
Glama
deenesjakoruzh
nezolder

Civil 3D MCP Server

by nezolder
OverviewSchemaSourceRelated ServersDiscussionsAdmin
Local

Civil 3D MCP Server — Dynamic Roslyn Fork

Ein MCP-Server, der KI-Assistenten ermöglicht, C#-Code direkt in Autodesk Civil 3D zu schreiben und auszuführen. Statt einer großen Menge fester Werkzeuge generiert die KI aufgabenspezifischen Code, der mit Zugriff auf die Civil-3D-API ausgeführt wird.

Projektumfang und Herkunft

Dieser Fork behält das dynamische Roslyn/C#-Ausführungsmodell und eine bewusst kleine öffentliche Oberfläche aus drei MCP-Werkzeugen bei. Die aktuelle Kompatibilitätsbasis ist Autodesk Civil 3D 2025, wobei sich die lokale Arbeit auf Zuverlässigkeit, Sicherheit, messbare Effizienz und wiederverwendbare Civil-3D-Skills konzentriert. Andere Civil-3D-Versionen können später durch separat verifizierte Kompatibilitätsarbeit hinzugefügt werden.

Das Projekt ist von barbosaihan/civil3d-mcp abgeleitet. SantosSjba/mcp-to-c3d wurde für ausgewählte, testgestützte Ideen evaluiert, während Sacred-G/Civil3D-mcp nur als architektonische Referenz verwendet wurde. Siehe PROVENANCE.md für die detaillierte Zuordnung und Lizenzgrenzen.

Dieses unabhängige Projekt ist weder mit Autodesk verbunden noch von Autodesk unterstützt. Autodesk-Assemblys und andere proprietäre Civil-3D-Dateien sind nicht enthalten.

Related MCP server: Civil 3D MCP Server

Architektur

┌─────────────────┐     stdio      ┌──────────────────┐     TCP/JSON-RPC    ┌──────────────────┐
│   AI Assistant   │ ◄────────────► │  MCP Server (TS) │ ◄──────────────────► │  Civil 3D Plugin │
│ (Claude, Cline)  │               │   3 meta-tools    │     port 8080       │  Roslyn Engine   │
└─────────────────┘               └──────────────────┘                      └──────────────────┘
                                         │                                         │
                                    Skills Library                           C# Code Execution
                                   (.skill.md files)                      (full Civil 3D API)

3 Meta-Tools

Tool

Zweck

Sicherheit

civil3d_execute

C#-Code mit Schreibzugriff ausführen; optionales Speichern nach dem Commit

⚠️ Ändert die Zeichnung

civil3d_query

C#-Code schreibgeschützt ausführen (kein Commit)

✅ Keine Nebenwirkungen

civil3d_skills

Code-Skill-Vorlagen durchsuchen/suchen/lesen; api_lookup durchsucht bereits geladene öffentliche Civil-3D-API-Metadaten

✅ Nur Metadaten

So funktioniert es

  1. KI liest einen Skill → Erhält eine dokumentierte C#-Codevorlage

  2. KI passt den Code an → Füllt Parameter aus, kombiniert Muster

  3. KI sendet Code → Über civil3d_execute oder civil3d_query

  4. Roslyn kompiliert und führt aus → Innerhalb von Civil 3D mit vollem API-Zugriff

  5. Ergebnisse werden als JSON zurückgegeben → Zurück an die KI

Beispielinteraktion

User: "What surfaces are in my drawing?"

AI: Uses civil3d_query with:
  var surfaces = new List<object>();
  foreach (ObjectId id in CivilDoc.GetSurfaceIds()) {
    var s = Transaction.GetObject(id, OpenMode.ForRead) as TinSurface;
    surfaces.Add(new { s.Name, s.Layer });
  }
  return surfaces;

Result: [{ "Name": "EG", "Layer": "C-TOPO-EG" }, ...]

Skills-Bibliothek

civil3d_skills unterstützt außerdem action: "api_lookup" für eine begrenzte, schreibgeschützte Suche nach öffentlichen Typ- und Membernamen/-signaturen aus bereits geladenen, auf der Whitelist stehenden Civil-3D-Host-Assemblys. Es lädt keine Assemblys, führt keinen C#-Code aus und greift nicht auf die aktive Zeichnung zu. Geben Sie eine Abfrage und optional eine Assembly, ein Namespace-Präfix und eine Ergebnisbegrenzung an.

Skills sind dokumentierte C#-Codevorlagen in skills/:

skills/
├── surfaces/           # Surface operations
├── alignments/         # Alignment + station/offset
├── points/             # COGO points
├── geometry/           # Lines, polylines, text
├── drawing/            # Drawing info
└── workflows/          # Complex multi-object operations

Skript-Globale

Code, der über civil3d_execute oder civil3d_query ausgeführt wird, hat Zugriff auf:

Global

Typ

Beschreibung

Document

Document

Aktives AutoCAD-Dokument

CivilDoc

CivilDocument

Aktives Civil-3D-Dokument

Database

Database

Dokumentdatenbank

Transaction

Transaction

Aktive Transaktion

Editor

Editor

Dokument-Editor

Alle Civil-3D-Namespaces werden automatisch importiert.

Das Committen der civil3d_execute-Transaktion ändert die geöffnete Zeichnung, schreibt die DWG-Datei aber nicht von sich aus auf die Festplatte. Setzen Sie saveDrawing: true, wenn die abgeschlossene Änderung ebenfalls gespeichert werden soll. Das Plugin speichert erst, nachdem die Skripttransaktion und die Dokumentsperre geschlossen sind; Skripte dürfen Database.SaveAs nicht selbst aufrufen oder QSAVE in die Warteschlange stellen. Die Speicheranfrage verwendet ein separates Standard-Timeout von 10 Minuten und wird nie automatisch wiederholt.

Einrichtung

1. MCP-Server erstellen

npm install && npm run build

2. Plugin erstellen

# Copy DLLs from Civil 3D to C_References/ (see C_References/README.md)
cd plugin/Civil3dMcpPlugin
dotnet build

3. In Civil 3D laden

NETLOAD → select Civil3dMcpPlugin.dll
C3DMCPSTATUS → verify running

4. KI konfigurieren

{
  "mcpServers": {
    "civil3d": {
      "command": "node",
      "args": ["/path/to/civil3d-mcp/build/index.js"]
    }
  }
}

Umgebungsvariablen

Variable

Standard

Beschreibung

CIVIL3D_HOST

localhost

Plugin-Host

CIVIL3D_PORT

8080

Plugin-Port

CIVIL3D_COMMAND_TIMEOUT

120000

Ausführungs-Timeout (ms)

CIVIL3D_SAVE_TIMEOUT

600000

Timeout für Ausführungsanfragen mit saveDrawing: true (ms)

LOG_LEVEL

info

Protokollebene

Benchmarking

Der hostunabhängige Recorder der Phase 2A, der Opt-in-Vertrag für interne Live-Traces der Phase 2A.1 und der schreibgeschützte Live-Runner der Phase 2A.2 sind in benchmark/README.md dokumentiert. Keiner fügt ein MCP-Werkzeug, eine Warteschlange oder einen Wiederholungsversuch hinzu; der 2A.2-Runner kann nur seine feste schreibgeschützte Abfrage aufrufen, wenn er explizit gestartet wird.

Strukturierte Fehler (Phase 2B.1)

civil3d_query und civil3d_execute behalten ihren bestehenden Textfehlerinhalt und isError: true bei und geben zusätzlich structuredContent mit dem Schema civil3d-mcp-error/v1 zurück. Die stabilen Fehlerfelder sind code, category, message, source, outcome und retryable. Ein Befehls-Timeout oder ein Verbindungsverlust nach dem Senden hat outcome: "unknown" und retryable: false; der Server wiederholt dies nie automatisch. Erfolgreiche Antworten und die öffentliche Oberfläche der drei Werkzeuge bleiben unverändert.

Privates TCP-Framing (Phase 2C.1)

Jede localhost-TCP-Verbindung überträgt eine UTF-8-JSON-RPC-Anfrage und eine Antwort. Jeder JSON-Body wird von LF gefolgt und ist auf 8 MiB begrenzt, gemessen als UTF-8-Bytes ohne das LF. Der Node-Client akzeptiert weiterhin die ungerahmte Antwort des vorherigen Plugins, wenn auf diesen vollständigen JSON-Body ein ordnungsgemäßes Schließen der Verbindung folgt. Überdimensionierte Anfragen werden abgelehnt, bevor sie geschrieben werden; überdimensionierte oder fehlerhafte Antworten und unterbrochene Verbindungen erzeugen nicht wiederholbare strukturierte Transportfehler. Wenn die Ausführung abgeschlossen war, das Plugin aber kein überdimensioniertes Ergebnis zurückgeben konnte, wird das Ergebnis als unknown gemeldet.

Operations-Überwachungsprotokollierung und Schreib-Idempotenz (Phase 2I.1 / 2I.2)

Bei der standardmäßigen Protokollebene info gibt jede akzeptierte civil3d_query- und civil3d_execute-Operation ein begrenztes stderr-Überwachungsereignis aus. Es enthält eine neue undurchsichtige Operations-ID, den Werkzeugnamen, SHA-256 und UTF-8-Bytelänge des C#-Quellcodes, den Erfolgs-/Fehlerstatus und die verstrichenen Millisekunden; Fehler fügen nur stabile code/category/source/outcome-Felder hinzu. Das Überwachungsereignis enthält niemals Aufrufercode, Beschreibung, Zeichnungsidentität, Ergebnis oder Fehlermeldung.

civil3d_execute akzeptiert außerdem einen optionalen undurchsichtigen idempotencyKey (1–128 ASCII-Buchstaben, Ziffern, ., _, :, -). In einer Plugin-Sitzung bindet es den Schlüssel an die UTF-8-C#-SHA-256, die normalisierte expectedDrawing-Identität und die saveDrawing-Auswahl. Ein Duplikat wird als in Bearbeitung, widersprüchlich oder bereits committet abgelehnt; committete Einträge behalten kein Ergebnis und Aufrufer müssen mit einer schreibgeschützten Abfrage abgleichen. Ein Speicherfehler tritt nach dem In-Memory-Schreib-Commit auf, daher wird der Schlüssel als abgeschlossen behalten, um eine versehentliche doppelte Änderung zu verhindern. Die Sitzung behält höchstens 256 abgeschlossene Schlüssel und entfernt deterministisch den ältesten. Dies fügt weder Persistenz noch automatische Wiederholung oder Exactly-Once-Semantik hinzu.

Sicherheit

Die Roslyn-Sandbox blockiert:

  • Prozessausführung (Process.Start)

  • Dateilöschung (File.Delete)

  • Netzwerkanfragen (HttpClient, Sockets)

  • Registrierungszugriff

  • Dynamisches Laden von Assemblys

Alle Civil-3D-API-Operationen sind erlaubt.

Diese Regex-Sandbox ist Verteidigung in der Tiefe, keine Vertrauensgrenze. Beide Code-Werkzeuge erhalten veränderbare Civil-3D- und AutoCAD-API-Objekte; civil3d_query überspringt den Transaktions-Commit des Hosts, kann aber nicht garantieren, dass beliebiger dynamischer C#-Code frei von Nebenwirkungen ist. Führen Sie nur vertrauenswürdigen, genehmigungspflichtigen Code aus. Loopback-TCP verhindert entfernten Netzwerkzugriff, authentifiziert aber keine anderen lokalen Prozesse.

Lizenz

MIT

Available Tools

3 tools
civil3d_executeA

Execute C# code in Civil 3D with write access. The code runs inside a committed transaction. Available globals: Document, CivilDoc, Database, Transaction, Editor. All Civil 3D namespaces are auto-imported. Return a value to get results back as JSON. Use this for operations that MODIFY the drawing (create, edit, delete objects). expectedDrawing must come from a prior read-only identity query. To persist the drawing file, set saveDrawing=true; do not call Database.SaveAs or queue QSAVE from the C# code.

ParametersJSON Schema
NameRequiredDescriptionDefault
codeYesC# code to execute. Has access to Document, CivilDoc, Database, Transaction, Editor. Example: var id = TinSurface.Create(Database, "MySurface"); return new { success = true };
descriptionNoOptional human-readable summary; excluded from operation audit logs.
saveDrawingNoWhen true, save the currently named DWG after the write transaction commits and wait for completion. Use this instead of Database.SaveAs or Document.SendStringToExecute("QSAVE") in code. An unsaved drawing must first be named in Civil 3D.
idempotencyKeyNoOptional opaque session key. Reuse it only to manually reconcile an uncertain outcome; use a new key for an intentional new write.
expectedDrawingYesExpected active drawing identity checked immediately before Civil API access.

TDQS

A4.7/5.0
Behavior4/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

With no annotations provided, the description carries the full burden and does well: it notes the committed transaction, available globals, JSON return, drawing identity check, and save workflow. It does not mention exception handling or failure rollback, but that is a minor gap for a code-execution tool.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness4/5

Is the description appropriately sized, front-loaded, and free of redundancy?

The description is compact and front-loaded with the core purpose, then elaborates on key parameters and constraints. It is not overly verbose, though bullet formatting could improve scannability; still, every sentence earns its place.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness5/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

For a complex code-execution tool with no output schema, the description explains available globals, return format, drawing identity requirements, save behavior, and sibling distinction. No critical operational detail is missing.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters5/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema coverage is 100%, yet the description adds significant context: expectedDrawing provenance and check timing, saveDrawing conditions (must be named), idempotencyKey purpose, and an example for code. It clearly enhances schema-only information.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

The description states a precise purpose: executing C# code with write access in Civil 3D. It explicitly scopes the tool to modifying the drawing ('Use this for operations that MODIFY the drawing'), which distinguishes it from the read-only sibling civil3d_query.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines5/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

It provides clear usage criteria: use for modifications, not for reads; requires expectedDrawing from a prior civil3d_query; and warns against calling Database.SaveAs or queuing QSAVE, directing the user to the saveDrawing parameter instead. This fully covers when and how to use it versus alternatives.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

civil3d_queryA

Execute C# code in Civil 3D in READ-ONLY mode (no changes saved). Available globals: Document, CivilDoc, Database, Transaction, Editor. All Civil 3D namespaces are auto-imported. Return a value to get results as JSON. Use this for querying data: listing objects, getting properties, analyzing surfaces, etc. Omit expectedDrawing only to bootstrap Database.Filename and Database.FingerprintGuid; otherwise supply it to guard the active drawing.

ParametersJSON Schema
NameRequiredDescriptionDefault
codeYesC# code to query data. Has access to Document, CivilDoc, Database, Transaction, Editor. Example: var surfaces = new List<object>(); foreach (ObjectId id in CivilDoc.GetSurfaceIds()) { var s = Transaction.GetObject(id, OpenMode.ForRead) as TinSurface; surfaces.Add(new { s.Name, s.Layer }); } return surfaces;
expectedDrawingNoExpected active drawing identity checked immediately before Civil API access.

TDQS

A4.3/5.0
Behavior4/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

With no annotations provided, the description carries the full burden and does substantial work: it discloses read-only semantics ('no changes saved'), available globals (Document, CivilDoc, Database, Transaction, Editor), auto-imported namespaces, the JSON return mechanism, and the expectedDrawing guard vs. bootstrap behavior. It does not cover error behavior for failed compilation or thrown exceptions at runtime, which is a notable gap for a code-execution tool, but the disclosed traits are rich.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness4/5

Is the description appropriately sized, front-loaded, and free of redundancy?

Three sentences, each earning its place: purpose/globals/return semantics, when-to-use, and the expectedDrawing rule. The first sentence is dense but not wasteful; the most critical differentiator (READ-ONLY) is front-loaded before supporting details.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness4/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

For a complex code-execution tool with no annotations and no output schema, the description covers the essentials: execution mode, environment globals, namespaces, return format, and the identity-guard parameter semantics. The main omissions are error/exception behavior and the exact failure mode when expectedDrawing mismatches, which an agent invoking arbitrary C# code would benefit from knowing.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters4/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema description coverage is 100%, so the baseline is 3. The description adds genuine value beyond the schema by explaining when to omit expectedDrawing entirely — 'Omit expectedDrawing only to bootstrap Database.Filename and Database.FingerprintGuid; otherwise supply it to guard the active drawing' — a semantic the schema's field descriptions do not convey.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

The description opens with a specific verb and resource: 'Execute C# code in Civil 3D in READ-ONLY mode (no changes saved).' It further scopes the tool with 'Use this for querying data: listing objects, getting properties, analyzing surfaces, etc.', which clearly differentiates it from the sibling civil3d_execute. An agent can tell immediately what this tool does and how it differs.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines4/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

'Use this for querying data' is an explicit when-to-use statement with concrete examples. The READ-ONLY framing implies that mutations belong to the sibling civil3d_execute, though it never names that alternative or states a when-not-to-use condition explicitly, so it stops short of a full 5.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

civil3d_skillsA

Browse and read Civil 3D code skills (documented C# code templates). Use 'list' to see available skills, 'search' to find by keyword, 'get' to read the full skill with code template, or 'api_lookup' to search public metadata from already-loaded Civil 3D host assemblies. Skills are pre-built C# patterns you can adapt and execute via civil3d_execute or civil3d_query.

ParametersJSON Schema
NameRequiredDescriptionDefault
limitNoMaximum list/search/api_lookup results to return (integer 1-50; default 20)
queryNoSearch query for 'search' or 'api_lookup' action
actionYeslist = browse skill metadata, search = find by keyword, get = read full skill, api_lookup = read-only public API metadata search
cursorNoOpaque nextCursor from a prior list/search call with the same filters
assemblyNoAllowlisted loaded host assembly filter for api_lookup
categoryNoFilter by category (surfaces, alignments, points, etc.)
namespaceNoNamespace prefix filter for api_lookup
skillNameNoSkill name for 'get' action

TDQS

A4.2/5.0
Behavior4/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

With no annotations, the description carries the behavioral disclosure burden. It clearly labels the tool as read-only ('Browse and read', 'read-only public API metadata search'), implying no state changes. It also notes that execution happens via sibling tools, which further clarifies that this tool itself does not modify anything. The absence of side-effect warnings is acceptable given the read-only framing.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness5/5

Is the description appropriately sized, front-loaded, and free of redundancy?

The description is three concise sentences, each adding critical information: the core purpose, the list of actions, and the relationship to sibling tools. It front-loads the main purpose and avoids redundancy or filler. Every sentence earns its place.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness4/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

For a read-only tool with four actions, the description is nearly complete. It explains the actions, implies the output (list of skills, search results, full skill content, API metadata), and points to the execution siblings. While it doesn't detail pagination or output structure, those are typically understood and the schema covers cursor details. The description is sufficient for an agent to call it correctly.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters3/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema description coverage is 100%, so all eight parameters have descriptions in the schema. The tool description adds contextual meaning (e.g., what 'get' does, that api_lookup is read-only) but does not explain parameter syntax or constraints beyond the schema. This meets the baseline of 3 but does not exceed it.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

The description opens with a specific verb-resource pair ('Browse and read Civil 3D code skills') and immediately enumerates the four supported actions (list, search, get, api_lookup). It also distinguishes itself from the sibling tools by noting that skills 'can be adapted and execute via civil3d_execute or civil3d_query.' This clearly sets its scope apart.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines4/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

The description explains the tool's role as a browsing/reading layer and explicitly points to the sibling tools for execution. It also differentiates between read actions (list/search/get) and the read-only metadata api_lookup. While it doesn't list explicit 'when not to use' scenarios, the purpose is clear enough for an agent to decide between this and its siblings.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

  1. 3 tool updatesv1.0.0
    • First observedcivil3d_execute
    • First observedcivil3d_query
    • First observedcivil3d_skills
Deploy Server
One endpoint for every MCP toolRayrun runs every MCP server behind one URL and holds the credentials. Each call is decided per client, tool, and argument, then recorded.ray.runAd

TDQS

A4.4/5.0

Scored across 3 tools

Disambiguation5/5

Each tool has a clear, non-overlapping purpose: execute for write operations, query for read-only operations, and skills for browsing code templates. The read/write distinction is explicitly stated, so agents should not confuse execute and query.

Naming Consistency4/5

All tools share the civil3d_ prefix and snake_case, but the suffixes mix verbs (execute, query) with a noun (skills), making it not a strictly consistent verb_noun pattern. The naming is still predictable and readable.

Tool Count5/5

Three tools is a well-scoped set for a server that provides arbitrary C# execution capabilities; each tool serves a distinct and necessary function. The count falls within the typical 3-15 range.

Completeness5/5

The combination of execute and query covers the full range of Civil 3D operations (create, edit, delete, query), and skills fills the learning gap. No obvious missing functionality for the stated purpose.

Maintenance

ActivityMaintained
ResponsivenessNo issues

Resources

Related MCP Connectors

Related MCP Servers

  • Civil 3D MCP Server

    DaniGhosy
    A
    license
    A
    quality
    C
    maintenance
    Enables AI assistants to interact with Autodesk Civil 3D through natural language, supporting tools for surfaces, alignments, profiles, corridors, pipe networks, COGO points, and AutoCAD geometry.
    9
    3
    MIT

  • Autodesk MCP Platform

    hjlrosales
    A
    license
    Not graded
    quality
    A
    maintenance
    Lets any MCP-compatible AI assistant read and edit Autodesk Civil 3D drawings through tools for alignments, surfaces, corridors, pipe networks, quantity takeoff, and cut/fill, using a local bridge plugin and named pipes.
    MIT