Tekla MCP Server connects MCP-compatible clients to Tekla Structures to automate modeling, selection, views, properties, components, operations, and drawings. The teknovizier/tekla_mcp_server repository describes a Python server based on FastMCP, with project resource discovery and a read-only mode. At the GitHub check on August 26, 2026, the repository was GPLv3 licensed, had 46 stars and 18 forks, and its documentation stated tests on Tekla 2022 and Tekla 2026.
For an architecture firm, Tekla MCP is not a replacement for a general BIM connector such as Revit MCP. It is mainly aimed at structural, precast concrete, steel, execution, and fabrication drawing teams that already work in Tekla Structures. Its value comes less from an AI demonstration effect than from its ability to read, select, filter, modify, and control very specific domain objects.
Table of Contents
- What Tekla MCP Server does
- Installation and architecture
- Domain capabilities
- Comparison with BIM and CAD MCPs
- Security, read-only mode, and limits
- Educasium position
- Sources and next reading
What Tekla MCP Server does
Key point: Tekla MCP Server exposes Tekla Structures as an environment that an AI assistant can query and operate. Its documentation distinguishes read-only tools, creative tools that add objects, and destructive tools that modify or delete existing data.
Tool families
The official reference groups tools into selection, components, properties, views, operations, drawings, and modeling. Selection can choose elements by filter, filter name, GUID, assembly, or main part. Properties can read objects, coordinates, bounding boxes, differences between elements, cut parts, and UDAs. Views cover temporary labels, zoom, redraw, isolate selected objects, hide selection, color selection, and apply view filters.
Components and modeling are more sensitive. The server can insert components, modify or remove existing components, place beams, columns, panels, slabs, and grids, and move, copy, or delete elements. Operations also include boolean cuts, attaching assemblies or rebars, macros, clash checks, and Tekla report creation.
Drawings and quality control
The drawings area is especially rich. The verified reference lists reading and modifying drawing properties, issuing or unissuing drawings, updating drawings, PDF printing, DWG, DXF, or DGN export, opening and closing drawings, reading views, reading annotations, finding unmarked objects, moving views, and aligning section views. It also includes checks for drawing collisions and unattached dimensions.
For a structural team, this may matter more than generating a beam. Drawings, marks, views, annotation collisions, and exports are frequent sources of repetitive work. An assistant that can inspect those elements under control can help prepare a review, but it should not decide alone to issue or modify production drawings.
Installation and architecture
Key point: Tekla MCP Server is a Python 3.11+ project that requires Tekla Structures and precise local configuration. It is not a one-click extension.
Requirements and dependencies
The quickstart requires Python 3.11 or higher and recommends uv pip install -r requirements.txt. The verified pyproject identifies the name tekla-mcp-server, version 0.2.3, and requires-python >=3.11. requirements.txt lists fastmcp 3.4.5, pathvalidate, pydantic, pythonnet, python-dotenv, sentence-transformers, and ezdxf.
pythonnet signals that the server talks to Tekla's .NET ecosystem. sentence-transformers is used for semantic attribute understanding. ezdxf is used in drawing workflows, especially to analyze exports. These dependencies explain why testing should be handled by someone able to diagnose Python, Windows paths, Tekla libraries, and MCP configuration.
Configuration files
The quickstart asks users to copy several sample files: settings.json, element_types.json, semantic_overrides.json, base_components.json, and report_properties.json. settings.json contains the Tekla path, context folder, read_only, excluded_tags, embedding settings, tolerances, and report or drawing settings. The server reads these files on startup, so every change requires a restart.
The TEKLA_MCP_LOG_LEVEL, TEKLA_MCP_LOG_FILE_PATH, and TEKLA_MCP_CONFIG_DIR environment variables set log level, log file, and configuration directory. The documentation emphasizes absolute paths for path variables. In practice, this architecture favors a documented pilot workstation rather than an improvised install across several machines.
Domain capabilities
Key point: Tekla MCP Server is built for structural and execution workflows, not just for drawing shapes. Its resources discover filters, components, catalogs, phases, grids, reports, macros, and context files.
Discovered resources
MCP resources include project://context, tekla://components, tekla://catalog/materials, tekla://catalog/rebars, tekla://macros, tekla://reports, tekla://element_types, tekla://filters/selection, tekla://filters/view, tekla://phases, tekla://grids, and tekla://connection_status. This gives the assistant more than commands. It can know filter names, available materials, rebar grades and sizes, phases, and project conventions if they are described in the context folder.
This matters in real production. An assistant that ignores numbering standards, Tekla classes, or local filter names will quickly produce unusable results. Project resources help align natural language with the team's conventions.
Components and semantic attributes
The README mentions a component handler system, with lifting anchors as an example: anchor selection based on element weight and placement calculated from the center of gravity. The configuration also describes base_components.json and custom properties exposed through tekla://components. The server can therefore be adapted to recurring components.
The configuration documentation describes semantic attribute mapping with a MiniLM model published on HuggingFace, teknovizier/minilm-tekla-attr-embed-v1, and an LLM fallback when confidence is uncertain. This helps when the user says a domain phrase instead of the exact Tekla attribute name. Caution remains necessary: every semantic mapping should be tested on a known component set before being allowed on a project.
Comparison with BIM and CAD MCPs
Key point: Tekla MCP Server is the most structure and fabrication drawing oriented connector in this series. It complements architecture BIM MCPs rather than replacing them.
| Option | Natural use | Main strength | Limit to watch |
|---|---|---|---|
| Tekla MCP Server | Structure, steel, precast, drawings, and components | Domain tools, project resources, read-only mode | Heavy configuration, destructive actions, GPLv3 |
| Revit MCP | General building BIM model | Architectural objects, levels, rooms, and extraction | BIM governance and Revit fork dependency |
| Civil 3D MCP | Site works, surveying, surfaces, and alignments | Roslyn C# inside Civil 3D, skill libraries | Code execution and infrastructure scope |
| CAD-MCP | DWG drafting through AutoCAD, GstarCAD, or ZWCAD | Multi-CAD Windows COM | Less rich in structural domain objects |
Difference from Revit MCP
Revit MCP acts on a more general building model: walls, floors, doors, rooms, grids, levels, and schedules. Tekla MCP Server goes deeper into structural objects, components, UDAs, assemblies, reinforcement, and drawings. The difference is therefore less about AI technology and more about domain scope.
For Educasium, this page should help architects who collaborate with structural offices and contractors. It can explain why automation that is useful in execution does not have the same perimeter as design-stage Revit automation. It should also remind readers that responsibility stays within the existing BIM and structural chain.
Difference from Civil 3D MCP
Civil 3D MCP relies on C# compiled through Roslyn inside Civil 3D. Tekla MCP Server instead offers a large surface of named and annotated tools, with read-only mode and exclusion tags. This granularity makes the perimeter easier to understand for a pilot.
The risk does not disappear. Destructive tools exist, macros can act broadly, and drawings can be updated or modified. But the documentation does provide governance levers: read_only, excluded_tags, category separation, project resources, and logs.
Security, read-only mode, and limits
Key point: read-only mode is the server's most important governance feature. It can expose inspection, selection, and navigation without presenting tools that write to the model.
read_only mode and tags
The configuration documentation says that read_only true exposes only tools annotated with readOnlyHint=true. Query, selection, and navigation tools remain available because they do not modify the model. Tools that write, whether additive or destructive, are hidden.
excluded_tags can also hide entire groups: resources, selection, view, properties, operations, components, drawings, or modeling. Both settings can be combined. For a firm or engineering office, the first reasonable scenario is therefore a read-only audit with logging before any modification command.
Technical limits
Documented compatibility is verified on Tekla 2022 and Tekla 2026. Other versions may work, but the compatibility page does not officially declare them supported. The quickstart also mentions possible issues such as a clr package conflict, a wrong tekla_path, missing Net48Runtime for Tekla 2026, missing JSON files, or an embedding model that needs to be downloaded.
These limits are concrete. A blocker can come from Python, Tekla, .NET, a Windows path, JSON configuration, or a HuggingFace model. An organization without a technical lead should account for that before promising an operational assistant.
Educasium position
Key point: Educasium can present Tekla MCP Server as a serious connector to study for structural workflows, but not as a solution already deployed by Educasium. The claims on this page come from verified GitHub sources, not a site report.
What we can say
We can say that the repository exists, is GPLv3, mentions FastMCP 3.0 in the README and fastmcp 3.4.5 in requirements.txt, is version 0.2.3 in pyproject, requires Python 3.11 or higher, and documents tests on Tekla 2022 and Tekla 2026. We can say that it provides read-only mode, exclusion tags, project resource discovery, and tools for selection, properties, views, components, operations, drawings, and modeling.
We should not say that Educasium installed it in a structural office, that it reduces drawing errors, or that it replaces an engineer's review. Those benefits would require measurements on a real project. Without a protocol, they must remain pilot hypotheses.
Hypothetical implementation
A hypothetical implementation would start with a Tekla test session, a non-client model, settings.json configured with read_only true, active logs, and a minimal context folder describing naming conventions. The first test would cover connection_status, filters, phases, grids, selection, properties, and bounding boxes. No modification tool would be exposed at this stage.
The second test would enable only one controlled category, for example read and export drawing tools, or modeling on a disposable model. The third would compare results with a human procedure: time, errors, changed elements, rollback ability, and log readability. Only after that framing can a team decide whether the connector deserves a wider pilot.
Sources and next reading
Key point: this page was written from GitHub sources checked on August 26, 2026. GitHub numbers will change, but GPLv3 status, configuration, read-only mode, and tool categories should remain the first verification points.
To continue, compare Tekla MCP Server with Revit MCP, Civil 3D MCP, and CAD-MCP. Tekla MCP Server is probably the most specialized of the four for structural and execution teams. It is also one of the connectors where access governance must be prepared before the demonstration, not after.