Ged Invest MCP
An MCP server for construction tools that calculates wall formwork material takeoffs from geometry inputs, enabling LLMs to generate accurate bills of materials.
README
Ged Invest MCP
An MCP (Model Context Protocol) server hosting a growing set of construction tools for Ged Invest. The first tool domain is wall formwork quantity takeoff; more tool domains can be added later as separate submodules.
Idea / architecture
Separation of concerns:
Photo / drawing / PDF
│
▼
[ LLM (ChatGPT / Claude) ] ← reads the geometry, builds structured data
│ { segments:[{length, height, corner}], system:"BAUTEKK" }
▼
[ Ged Invest MCP (Python) ] ← deterministic calculators (tools)
│ panel selection + corners + hardware + aggregation
▼
[ Result: BOM ] → panels, corners, ties, props, m², timber
- The LLM is the "eyes": it reads an image and turns it into numbers.
- The MCP is the "calculator": repeatable, no guessing. The same geometry always yields the same result - unlike LLM arithmetic.
Tool domains
| Domain | Status | Tools |
|---|---|---|
formwork |
ready | list_formwork_systems, formwork_catalog, count_formwork, concrete_pressure_check |
| (future) | planned | additional construction tools register the same way |
Formwork systems
| System | Max pressure | Catalog |
|---|---|---|
| BAUTEKK | 40 kN/m² | verified against DTR BauTekk 04/2022 + project PDF (R01) |
| BAUFRAME | 60 kN/m² | default values — to be verified |
| BAUSCHAL | 80 kN/m² | default values — to be verified |
BAUTEKK data taken from the manufacturer DTR: panel widths (10/20/25/30/45/50/60/75/90 cm, 70 cm as VT), three heights (90/120/150 cm), real article numbers for every size, outer/inner corners per height, and the connector rule (5/4/3 connectors per joint for 150/120/90 cm plates). Tie and prop counts remain design-dependent estimates.
BAUFRAME/BAUSCHAL catalogs contain sensible starting values. Confirm panel sizes and article numbers against the official BAUKRANE catalog before production use (
src/ged_invest_mcp/formwork/catalog.py).
Install
Requires Python ≥ 3.10.
python3.13 -m venv .venv
source .venv/bin/activate
pip install -e .
Run
The server supports two transports from a single entry point.
stdio (local — Claude Desktop / Cursor)
python -m ged_invest_mcp.server # stdio by default
ged-invest-mcp
In this mode you host no network server — the client (Claude/Cursor) launches the Python process itself and talks to it over stdin/stdout. When the client closes, the process ends. No ports, no hosting.
HTTP (remote — ChatGPT / connectors)
python -m ged_invest_mcp.server --http # http://0.0.0.0:8000/mcp
python -m ged_invest_mcp.server --http --port 9000
python -m ged_invest_mcp.server --transport sse # legacy SSE transport
ChatGPT cannot launch a local process, so it needs an MCP exposed over HTTP at a
URL. Expose a local server via a tunnel (e.g. ngrok http 8000) or deploy it,
then add https://.../mcp as a custom connector.
count_formwork — two ways to pass geometry
A) Segment list (you know wall lengths and corners):
{
"system": "BAUTEKK",
"segments": [
{"length": 465, "height": 150, "corner_at_end": "outer", "label": "A"},
{"length": 465, "height": 150, "corner_at_end": "outer", "label": "B"}
]
}
B) Polygon (outline coordinates — corners detected automatically):
{
"system": "BAUTEKK",
"polygon": [[0,0],[840,0],[840,700],[0,700]],
"height": 150
}
Client configuration
Claude Desktop / Cursor (mcp.json)
{
"mcpServers": {
"ged-invest": {
"command": "/ABSOLUTE/PATH/bud-ged-mcp/.venv/bin/python",
"args": ["-m", "ged_invest_mcp.server"]
}
}
}
Method (formwork, in short)
- Wall outline → segments between corners (outer / inner / none for collinear).
- Segment length → panel width selection (dynamic programming, min. timber).
- Height → panels are stacked to reach the wall height; the top course may
stand above the pour line (reported as
top_overshoot_cm, panels are not cut). - Panels = columns × courses × faces (default 2 = wall centerlines; pass
faces=1if the input already lists both faces separately). - Corners: one corner element per course.
- Hardware: full family from the catalog (ties, nuts, cones, plugs, spacer tubes, connectors, pins, wedges, corner tensioners, props with heads/feet). Coefficients are approximate and should be calibrated against the DTR.
- Horizontal leftover that does not fit panels → timber infill [lm].
Auditable output
count_formwork returns, besides the BOM:
input_echo— the geometry actually used (polygon-derived side lengths, corner types and winding), so you can verify the LLM read the drawing correctly;assumptions— faces, corner rule, spacings, panel heights, policies;- per-segment
panels_horizontal_per_face/courses_vertical_per_facelayout; reconciliation— geometry area vs panel area and waste %;units,catalog_version,warnings,disclaimer.
This is an engineering estimator — the result supports quoting/ordering and must be approved by the site manager (per the project notes).
concrete_pressure_check — fresh concrete pressure (DIN 18218)
Computes the maximum characteristic lateral fresh concrete pressure per
DIN 18218:2010-01 from the pouring (rise) rate v, consistency class
(F1–F6, SCC) and setting time tE, caps it by the hydrostatic pressure
γc·H, and compares it to the formwork system's limit:
characteristic_pressure_kn_m2,hydrostatic_pressure_kn_m2,design_pressure_kn_m2,governing;within_limitandmax_pouring_rate_for_limit_m_per_h(the fastest safe rate);- pass
system(e.g.BAUTEKK) to auto-fill the limit (40 kN/m²) and the DTR process cap (2 m/h).
Simplified 15 °C reference model (no temperature/admixture corrections). The real pressure and safe pouring rate remain the site manager's responsibility.
Adding a new tool domain
- Create
src/ged_invest_mcp/<domain>/with aregister(mcp)function. - Import and call it in
server.py.
Tests
python -m pytest -q
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.