OpenTakeoff MCP Server

Construction takeoff for AI agents: load plans, set scale, measure, count, export with provenance.

Publisher claimed. No tool list reported, and Pod has not connected to this server.

Status

Pod has not dialled OpenTakeoff yet, so everything on this page is what its publisher reported rather than what we observed. Registries describe servers; they do not connect to them. Until a check runs, treat the tool list below as a claim.

Connect

Published as opentakeoff-mcp on npm. Runs locally.

Known issues

20 problems reported by people outside the maintainer team. Issues filed by the project's own owners, members and collaborators are excluded — those are release checklists and internal refactors, not things that will go wrong for you. Showing 12.

Most discussed

RFC: the sheet graph — resolve room tags, schedules, legends, and detail callouts across a plan set

A single sheet never carries the answer

A-201 says room 102 — OFFICE. The finish schedule on A-001 says 102 → CP-1 / RB-1. The legend says CP-1 = carpet tile 24×24. Wall section 3/A-601 says the partition is 9'-0", which is what turns 128 linear feet of wall into a real RB-1 base quantity.

OpenTakeoff gives an agent read_sheet_text — positioned text, one sheet at a time — and nothing else. Every agent that has driven this engine has re-derived the entire cross-reference st

Read the thread · 2026-07-22 · closed · 8 comments

RFC: read the PDF layer tree — Optional Content Groups as first-class geometry filters

The heuristics exist to recover information the file already states

classifyHatchSegs in web/src/lib/oneclick.ts is a wall of tuned constants — HATCH_MIN_RUN, pitch-regularity bands, SPAN_PROTECT_RATIO, the filled-not-stroked poché exemption added after the VA plan drew its walls as solid shapes — and its entire job is to guess which segments are boundary and which are pattern.

A construction document exported from Revit, AutoCAD, or Archicad usually says so outright. PDF Optional C

Read the thread · 2026-07-22 · closed · 4 comments

MCP: a versioned computed-report export — opentakeoff.report.v1

Two downstream consumers have now built pricing tools on top of OpenTakeoff exports, and both hit the same gap: there is no first-class, versioned export of the computed takeoff — per-condition totals plus the materials buy list.

What exists today

`export_takeoff` emits `opentakeoff.takeoff_canvas.v1` — the raw annotations payload, exactly what the app autosaves. Perfect for round-tripping into the canvas; wrong shape for a consumer that wants totals.
`takeoff_summary` computes

Read the thread · 2026-07-29 · closed · 1 comment

MCP: waste is real math the agent can't reach — no tool sets a condition's waste_pct

The gap

takeoff_summary emits waste_pct and waste-adjusted *_net order quantities per condition — net figures are a first-class output of the server. But no tool can set waste: conditions minted through one_click / measure_polygon / measure_line / edit_materials start at waste_pct: 0 and nothing in the tool surface changes it (grep -i waste mcp/src/tools.ts matches only the takeoff_summary description). So for every agent-driven takeoff, net === gross.

Surfaced concretel

Read the thread · 2026-07-29 · closed · 1 comment

MCP: computed-report export — opentakeoff.report.v1 (totals + materials buy list), not just the raw canvas payload

The gap

The MCP server has two ways to get numbers out of a session:

takeoff_summary — per-condition quantities (floor/wall/border SF, LF, EA, SY, gross and *_net) plus grand totals. By design it strips materials for a compact quantities-only reply (mcp/src/outputs.ts, the edit_materials description).
export_takeoff — the exact opentakeoff.takeoff_canvas.v1 autosave payload. Materials appear here as config rows ({name, per, basis, unit, round}), not as computed ord

Read the thread · 2026-07-29 · closed · 1 comment

Most recent

MCP: cut_out — a real hole, reconciled, the way the canvas cuts one

An estimator has Cut Outs (#137): draw a hole inside a parent shape and the engine subtracts it for real — N holes per parent, overlap never double-deducts, the report and legend read the reconciled number, and the whole thing is one undo step.

An agent has role: "deduct". That commits an independent overlay shape: nothing links it to a parent, accumulateRole treats it as its own scope, and the marked set paints a decal rather than a hole. The two paths disagree about what a deduction *is

Read the thread · 2026-08-04 · closed · 0 comments

MCP: raster one_click for scanned sheets — wire the fallback the canvas already has

The canvas one-click falls back to a raster mask on scanned sheets (web/src/lib/rastermask.ts: Bradley adaptive threshold, vector-always-wins, raster_traced provenance). The MCP server refuses instead: 'This sheet has no vector linework… raster fallback not yet available in the MCP server.' The engine code exists and is pure; this is wiring, provenance passthrough, and a fixture. (Filed for completeness — vector-plan users never hit it, scanned-set users hit it immediately.)