Skip to main content
Glama

dcc-mcp-paraview

A typed MCP adapter for a persistent, isolated ParaView pipeline. It creates and edits spheres, clips with planes, reads single-file VTI/VTP/VTU datasets, extracts contours, exports verified datasets, saves native PVSM states, and supports display-backed PNG previews.

Status: local Core/server 0.20.41 upgrade candidate; native requalification is pending. Historical Linux acceptance used ParaView 5.13.2, Core/server 0.20.39, Python 3.12.14 and official MCP SDK 1.30.0. Those results do not qualify this upgrade. No published release is claimed. Existing ParaView GUI sessions are untouched.

Start

Install ParaView from its official distribution or your supported system package source. Keep its pvpython interpreter separate from the adapter's Python environment.

python -m venv .venv
.venv/bin/python -m pip install -e '.[dev]'
mkdir -p artifacts
.venv/bin/dcc-mcp-paraview --workspace "$PWD/artifacts" --pvpython /path/to/pvpython

The process prints its direct loopback mcp_url and registry instance_id. The default port is ephemeral; use --port or DCC_MCP_PARAVIEW_PORT when a fixed direct endpoint is needed. DCC_MCP_PARAVIEW_EXECUTABLE also selects pvpython. Ctrl+C shuts down the endpoint, registry entry, pump and owned host process group. Startup enforces the candidate Core/server 0.20.41 pins and ParaView 5.13.2 host gate. The Linux-only product gate is unchanged. Native pipeline, real HTTP lifecycle/cancellation and display-backed render acceptance must be repeated on this runtime before qualification. See the upgrade record.

The adapter defaults Core's gateway port to zero, so it does not elect a gateway or open a LAN listener. A studio can explicitly configure Core gateway options through the Python server constructor after reviewing deployment policy.

Related MCP server: linux-computer-mcp

MCP workflow

  1. Initialize the direct MCP endpoint

  2. Call search_skills with {"query":"paraview"}

  3. Call load_skill with {"skill_name":"paraview-pipeline"}

  4. Follow all pages of tools/list; call the returned tool names

  5. For each async result, poll its existing Core job_id with jobs_get_status to a terminal state

On the historically tested Core 0.20.39, tools/list exposes short names such as create_sphere; discovery also supplies namespaced slugs. Do not guess or replay a mutation after a timeout.

Tool

Purpose

get_adapter_info

Pure host-independent limits and formats

inspect_pipeline

Source geometry, point arrays and exact host-thread identity

create_sphere, edit_sphere

Bounded sphere parameters, verified after writes

clip_plane

Native plane clip of an existing input

slice_plane

Bounded true polygonal section of an existing input

inspect_presentation

Read existing cameras, displays, palettes and legends

capture_current_view

Export the configured view only if presentation stays unchanged

open_dataset

Local single-file VTI, VTP or VTU import

contour

At most eight isosurfaces of a named point scalar

export_dataset

VTI/VTP/VTU export with native geometry/array readback

save_state, reopen_state

Native PVSM save and trusted same-session reopen

render_preview

PNG, scalar palette/range, camera and background

See the installable skill and exact tool schemas.

Deliberate limits

  • The candidate is Linux-only with Python 3.10–3.12 for the sidecar, independently of vendor Python. Core/server 0.20.39 evidence is historical; validation separates the measured 0.20.41 commit and its passing CI from later review fixes awaiting fresh qualification. Windows pipe handling is not implemented; Python 3.7 LTS acceptance is not claimed

  • No attachment to existing GUI pipelines, arbitrary Python, programmable filters, remote servers, distributed MPI, time-series readers or general PVSM imports

  • Native state can contain executable filters. reopen_state accepts only unchanged files saved by the same live adapter session. Saved PVSM remains a native editable artifact for ParaView; after an adapter restart use verified datasets or open the native state yourself in ParaView

  • All file operations are confined to one operator-owned workspace. Inputs are limited to 64 MiB and pipelines to 32 sources. Outputs never overwrite; all symlink path components are rejected. This is a single-user tool boundary, not an OS sandbox for hostile concurrent filesystem changes or hostile native file parser inputs

  • Rendering requires a working DISPLAY; use the opt-in isolated-display acceptance test. Native output dimensions, decoded pixels, camera, background and scalar transfer function are verified. EGL/OSMesa auto-detection is intentionally absent. A defined but broken display can still crash the owned host

  • Native calls are monolithic. Timeout/cancellation terminates the owned host, discards its unsaved state and permanently closes that session. A file operation already published before interruption may remain. No rollback or cross-restart job recovery is claimed

Architecture · Installation and removal · Validation evidence

The slice and presentation candidate adds native plane sections and capture that detects first-render camera resets. The development record retains its earlier pending-qualification wording.

Selected Slice qualification

The later source-author report for one installed-wheel Linux native five-view test is recorded in the selected qualification. It supplements the frozen candidate's development record. Publication checks are separate host-free runs and exact-head CI; they do not rerun that native or GUI evidence or qualify every possible presentation property.

The separate line and legend controls candidate adds optional bounded placement, text/font and line-width controls. Its selected qualification records later source-author native evidence, source identity and the correction of legend-thickness units. Publication validation runs host-free checks separately.

Related MCP Connectors

Related MCP Servers

  • A
    license
    Not graded
    quality
    D
    maintenance
    Integrates multimodal large language models with ParaView to enable the creation and manipulation of scientific visualizations using natural language and visual inputs. It features visual feedback capabilities for iterative refinement, making advanced visualization workflows accessible through intelligent automation.
    69
    BSD 3-Clause
  • A
    license
    Not graded
    quality
    B
    maintenance
    Enables AI agents to work in a persistent, isolated Linux workspace with file management, Bash execution, SSH/SFTP access, and durable storage while keeping workloads contained from the host and private networks.
    MIT