TD Builder MCP server
Commit Oct 29, 2026Enables LLMs to generate, validate, and edit TouchDesigner networks locally using a key-free knowledge base and live editing tools.
Also known as TrueFiasco/TD_Builder_alpha, td-builder-alpha
- Stars
- 1
- Tools
- 1
- Downloads, 30 days
- —
- Latest version
- —
- Last commit
- Oct 29, 2026
- License
- AGPL-3.0
What is TD Builder MCP?
TD Builder enables LLMs to generate, validate, and edit TouchDesigner networks locally using a key-free knowledge base and live editing tools. It exposes 1 tool, such as td_build_project.
Summary generated from the project's README
About TD Builder MCP server
TD_Builder_alpha provides a local environment for AI clients to generate, validate, and edit TouchDesigner networks without requiring any API keys. The system operates through two distinct MCP servers to manage context efficiently. The first server, td-builder, offers 18 tools for offline tasks such as KB search, project building, and operator validation. The second server, td-builder-live, provides 22 tools specifically for interacting with a running TouchDesigner instance, including node CRUD operations and system introspection.
Users run this system locally using Python 3.10 through 3.13. It utilizes a local knowledge base containing operator specifications and wiki guides, which is fetched during installation. To maintain privacy and independence from cloud services, the system uses a local embedding model for semantic search. This setup is designed for developers working with TouchDesigner 2025 or newer who need an offline-capable way to build complex operator networks and manage project files through an LLM interface.
What you can do with it
- 1Live environment manipulationAn AI client can perform real-time edits and inspections on a currently running TouchDesigner session.
Written from the project's README. Every use case is checked against a quote from it.
TD Builder MCP tools 1
Detected from the READMEtd_build_project
| td_build_project | — |
Automated checks
Ran Nov 14, 2026Full report →Not verified yet. A server needs a score of 80, no serious finding, and enough data for us to check. How checks work · What to fix for more points
- Source repository linked and reachabletruefiasco/td_builder_alpha is public and reachablegithub.com/truefiasco/td_builder_alpha ↗Pass
- Package names this repository as its sourceNot checked
- Build provenance (Sigstore attestation)Not checked
- Open source license (OSI)AGPL-3.0spdx.org/licenses/AGPL-3.0.html ↗Pass
- Commit or release in the last yearLast commit or release 2026-10-29github.com/truefiasco/td_builder_alpha ↗Pass
- OpenSSF ScorecardNot checked
- Package name not close to a popular packageNot checked
- No critical advisory in the package or its dependenciesNot checked
- No hidden instructions to the model1 tool description and the README read; no override phrases, credential paths or hidden charactersPass
- Destructive tools are labelledNo tool named like delete, drop or execPass
- Remote endpoint answers the MCP handshakeNot checked
README
View on GitHubTD Builder — v0.2.1
A key-free Python system for AI-driven generation, validation, and conversion of TouchDesigner networks, exposed to LLM clients as two MCP servers, plus live TouchDesigner editing/feedback tools. Everything runs locally — no API key required.
Folder layout
| Folder | What's in it |
|---|---|
MCP/ |
The two MCP servers + all the code they run. server.py (offline) and live_server.py (live-TD) are the entry points; server_core/ is the server brain, engine/ is the TD-file engine, live_client/ holds the live-TD client, td-webserver/ is the TouchDesigner-side .tox asset. |
KB/ |
The knowledge base (operator specs, graph, vector store, wiki guides). Fetched/placed on install — see KB/README.md. |
Agents/ |
Expert prompts (experts/), learned expertise (expertise/), skills (td-builder-howto/, td_network_analysis/), and strategy notes (Strategies/). The server reads experts/expertise from here. |
Tools/ |
LLM-facing documentation of every tool (TOOLS.md + the KB tools/, Live tools/, offline Builder tools/, Other/ categories) and the offline CLI launchers. |
Config/ |
User configuration: .env.template, search_config.json, and SETTINGS.md. |
LLM/Pre-Prompts/ |
Reusable pre-prompts and when to use them. |
scripts/ |
fetch_vector_db.py (downloads the KB bundle), check_deps.py (verifies your install). |
tests/ |
Acceptance + smoke gate so you can confirm a working install. |
The two MCP servers
| Server | Register as | Tools | When |
|---|---|---|---|
MCP/server.py |
td-builder |
18 key-free tools — KB search, td_validate/td_convert/td_build_project/td_build_status, expand_toe_file, register_component, get_expert_prompt, get_server_info |
always |
MCP/live_server.py |
td-builder-live |
22 live tools — capture / node CRUD / introspection of a running TouchDesigner | only when TouchDesigner is open |
Keeping the live tools in a separate server means offline sessions don't carry their ~22 tool schemas in the model's context. See MCP/README.md for the client config.
Quick start (Windows + PowerShell)
Prerequisites: Python 3.10–3.13 (3.11 recommended), TouchDesigner 2025+ (only for the live tools + final
.toecollapse). See PREREQUISITES.md.
# 1. Create a venv on Python 3.11
py -3.11 -m venv .venv
.\.venv\Scripts\Activate.ps1
# 2. Install (single editable install + dev/test extras)
pip install -e ".[dev]"
# 3. Download the knowledge base (~108 MB zip) — public HTTPS, no GitHub account needed
python scripts\fetch_vector_db.py
# 4. Verify your install
python scripts\check_deps.py # expect all green
# 5. Register the MCP server(s) with your client — see MCP/README.md
# IMPORTANT: point the client's "command" at THIS venv's python
# (<ROOT>\.venv\Scripts\python.exe; macOS/Linux: <ROOT>/.venv/bin/python) —
# a bare "python" is the system interpreter and dies with ImportError.
# Offline: .venv\Scripts\python.exe MCP\server.py (td-builder, 18 tools)
# Live: .venv\Scripts\python.exe MCP\live_server.py (td-builder-live, 22 tools — only with TD open)
First run: the first KB-dependent call loads the knowledge base (~180 MB) + the local embedding model + the bundled reranker (one-time, typically 5–10+ min), then every call is fast.
Modes
There is one mode: key-free. KB semantic search uses a local embedding model
(all-MiniLM-L6-v2) — no Voyage/OpenAI/Anthropic key, no cloud calls. (Agent-spawning and the
multi-agent strategy runner were removed in this release; the experts ship as prompts you load via
get_expert_prompt.)
Measured, not promised
The KB and builder are gated by a published eval harness (eval/) — every release must pass:
- Build correctness: offline-generated operators re-expanded through TD's own
toeexpandand diffed against live-TD ground truth — 636/636 operator tokens exact (verified by importing offline.toxfiles into a running TouchDesigner). - Retrieval quality: labeled benchmark (recall@k / MRR / nDCG) with a frozen trend gate, held-out paraphrases against overfitting, and a coverage tier spanning ≥90% of operators in every family (all 663 KB entries enumerated). Hybrid retrieval (dense + BM25 + reranker) took the aggregate score from 0.86 → 0.93 over dense-only.
- Name integrity: zero retokenized operator names in returned chunks (was 294 pre-redesign).
Run the gate
py -3.11 -m pytest tests\acceptance tests\measure -q # ~22 checks; live tests need TD open
Known limitations
- BASIC-mode
.toebuilding (from-scratch networks) emits parameter-format warnings; the LOSSLESS round-trip path (the CLIs /td_fixture_pipeline) is solid. td_build_project(palette=…)is a known-broken path (deferred to a later release).- See docs/KNOWN_ISSUES.md.
License
TD Builder is licensed under the GNU Affero General Public License v3.0 or later (AGPL-3.0-or-later) — see LICENSE. Note AGPL §13 (Remote Network Interaction): if you run a modified version of these MCP servers for others to interact with over a network, you must offer those users the Corresponding Source of your modified version.
TD Builder MCP questions
What tools does TD Builder MCP provide?
It exposes 1 tool, including td_build_project.
What are the system requirements for running the live tools?
You need Python 3.10–3.13 and TouchDesigner 2025 or newer.