Direct Use Exposure Mcp

Direct-Use Exposure MCP

[](https://github.com/ToxMCP/direct-use-exposure-mcp/actions/workflows/ci.yml) [](./LICENSE) [](https://github.com/ToxMCP/direct-use-exposure-mcp/releases) [](#release-verification) [](https://www.python.org/)

> Part of ToxMCP Suite → https://github.com/ToxMCP/toxmcp

Public MCP server for auditable deterministic exposure screening in regulatory and scientific AI workflows. It turns product-use assumptions into reproducible dermal, direct-use/incidental oral, inhalation, jurisdictional-comparison, and aggregate external-dose scenarios with explicit provenance, limitations, quality flags, and fit-for-purpose guidance. It also exports structured evidence objects and PBPK-ready handoff payloads without taking over PBPK execution, WoE synthesis, BER, PoD derivation, or final risk decisions.

Use it when you need:

• a stable, inspectable external-dose object rather than free-text exposure reasoning
• governed defaults and explicit scenario assumptions that can survive review
• deterministic cross-jurisdiction comparisons without black-box probabilistic overclaiming
• clean handoffs into PBPK or broader ToxMCP orchestration layers

Architecture

flowchart LR
    subgraph Clients["Clients and Orchestrators"]
        Codex["Codex CLI / Desktop"]
        Scripts["Scripts / notebooks"]
        Other["Other MCP-aware agents"]
    end

    subgraph MCP["FastMCP Service"]
        Server["Tool and resource surface"]
        Contracts["Schemas, examples,\ncontract manifest"]
        Prompts["Refinement and\nhandoff prompts"]
    end

    subgraph Engine["Scenario Engine"]
        Runtime["Deterministic runtime"]
        Screening["Dermal / direct-use oral screening plugin"]
        Inhalation["Inhalation screening plugin"]
        Aggregate["Aggregate/co-use summary"]
    end

    subgraph Evidence["Scientific Control Layer"]
        Defaults["Versioned defaults packs"]
        Provenance["Assumption ledger,\nprovenance, quality flags"]
        Review["Release readiness,\nsecurity, provenance review"]
    end

    subgraph Downstream["Suite Handoffs"]
        PBPK["PBPK MCP import bundle"]
        EvidenceBundle["Evidence export bundle"]
        Refinement["Refinement export bundle"]
    end

    Clients --> Server
    Server --> Contracts
    Server --> Prompts
    Server --> Runtime
    Runtime --> Screening
    Runtime --> Inhalation
    Runtime --> Aggregate
    Runtime --> Defaults
    Runtime --> Provenance
    Server --> Review
    Server --> PBPK
    Server --> EvidenceBundle
    Server --> Refinement

The core engine is intentionally narrow, even though the released MCP also publishes bounded worker, exchange, and validation surfaces:

• Direct-Use Exposure MCP owns external-dose construction only.
• Current suite interoperability is with CompTox MCP, ADMETlab MCP, AOP MCP,

O-QT MCP, and PBPK MCP.

• PBPK MCP owns kinetic translation and internal-dose interpretation.
• Defaults, assumptions, provenance, and limitations are first-class outputs, not hidden internals.
• Dietary and fate seams are still explicit: medicinal or product-centric oral regimens stay

here, while food-mediated intake and multimedia release/concentration workflows remain separate future module boundaries.

For a one-page maturity framing of the full released surface, see [docs/capabilitymaturitymatrix.md](./docs/capabilitymaturitymatrix.md). For a suite-level map of sibling services and shared handoff contracts, see [docs/toxmcpsuiteindex.md](./docs/toxmcpsuiteindex.md).

ToxMCP suite fit

This repo is the public exposure-construction module inside the broader ToxMCP Suite. The current public module map is:

| Module | Role in the suite | Relationship to this repo | | --- | --- | --- | | Direct-Use Exposure MCP | Direct-use and near-field external-dose construction | This repo | | CompTox MCP | Identity, hazard, and EPA CompTox-backed enrichment | Upstream evidence/context source | | ADMETlab MCP | Rapid ADMET prediction utilities | Adjacent screening module | | AOP MCP | Mechanistic pathway and AOP workflows | Adjacent mechanistic module | | O-QT MCP | OECD QSAR Toolbox workflows and reporting | Adjacent modeling module | | PBPK MCP | Internal-dose and TK simulation | Downstream handoff target |

Two additional seams are already documented here but are not current public modules in the umbrella repo:

• Fate MCP: environmental release, multimedia transfer, and concentration surfaces
• Dietary MCP: food-mediated intake, commodity residue, and dietary oral workflows

That distinction matters for the README and contract story: this MCP should read as one module in a growing suite, not as the whole ToxMCP platform.

What's in v0.2.0

• Deterministic dermal plus direct-use/incidental oral screening scenario construction
• Deterministic inhalation screening with room-volume, ventilation, saturation-cap, and deposition semantics
• Tier A uncertainty registers, deterministic sensitivity ranking, and dependency metadata
• Tier B deterministic scenario envelopes, archetype-library sets, and bounded parameter propagation
• Tier C single-driver probability bounds plus coupled scenario-package probability profiles without Monte Carlo overclaiming
• Machine-actionable Tier 1 inhalation upgrade advisories plus packaged airflow, particle, and product-family screening profiles
• Curated RIVM-backed dermal contact defaults plus governed spray airborne-fraction defaults for personal-care and household-cleaner contexts
• Evidence reconciliation across CompTox, SCCS, SCCS opinions, CosIng, ConsExpo, nano/micro guidance, and reviewed user-supplied product-use records
• Particle-aware evidence lanes for EU cosmetic nanomaterials, synthetic polymer microparticles, and non-plastic micro/nanoparticles
• Integrated evidence-to-scenario-to-PBPK workflow execution as one audited MCP response
• External-dose aggregate summaries plus opt-in route-bioavailability-adjusted internal-equivalent screening totals
• Scenario comparison, refinement deltas, evidence export, and refinement-bundle export
• PBPK scenario export plus exact external-import payload packaging, with optional transient inhalation concentration profiles
• Published JSON schemas, examples, contract manifest, shared cross-MCP contracts, and release metadata
• Release-readiness, verification, result-status, troubleshooting, provenance, and scientific-boundary resources
• Source-backed herbal/TCM/supplement routing plus governed medicinal oral, supplement, topical spray, and topical patch anchors
• Worker-task routing, Tier 2 bridge/export support, governed ART external exchange, and bounded worker dermal execution
• Validation dossier, validation coverage report, executable reference bands, executable time-series packs, and showcase goldset resources

Why this project exists

Exposure information is often the weakest structured input in early NGRA orchestration: there may be CompTox context, product-use hints in prompts, or local refinement notes, but not a stable, auditable external-dose object that downstream systems can trust.

Direct-Use Exposure MCP gives the suite a dedicated exposure layer that is:

• deterministic-first for transparent screening use
• MCP-native with typed tools, resources, prompts, schemas, and examples
• auditable through assumption records, defaults versioning, provenance, and quality flags
• bounded so it complements PBPK and adjacent review/orchestration layers instead of overlapping them

Who this is for

• Regulatory toxicologists and exposure assessors who need scenario outputs that remain reviewable and traceable.
• Product safety, stewardship, and EHS teams who need consistent screening scenarios before higher-tier assessment.
• Consultants and dossier-support teams who need reproducible exposure objects for briefings, memos, and handoff packages.
• Scientific AI and MCP builders who need a trustworthy exposure module inside a broader governed workflow.

Capability maturity

The repo now has a broader released surface than the early "small deterministic builder" story implied. The cleanest way to read it is:

• core deterministic exposure engine: benchmark-regressed external-dose construction
• evidence reconciliation and integrated workflow: external-normalized orchestration helpers
• worker inhalation and dermal: bounded extension layers with explicit solver limits
• validation and release resources: first-class trust and governance surface

The detailed maturity matrix is in [docs/capabilitymaturitymatrix.md](./docs/capabilitymaturitymatrix.md).

Feature snapshot

| Capability | Description | | --- | --- | | 🧪 Screening scenarios | Builds route-specific external-dose scenarios for dermal, direct-use/incidental oral, and inhalation screening use cases, with bounded volatility saturation caps, first-order deposition sinks, and explicit extrathoracic swallowed-mass handoff metrics on spray inhalation branches where applicable. | | 📊 Tier A uncertainty diagnostics | Publishes qualitative uncertainty registers, one-at-a-time sensitivity ranking, dependency metadata, and validation posture on each scenario. | | 📦 Tier B deterministic envelopes | Builds named archetype envelopes with bounded min/median/max outputs and explicit driver attribution without probabilistic overclaiming. | | 🗂️ Tier B archetype library | Publishes governed packaged archetype sets, including Tier 1 inhalation request templates where near-field screening is part of the intended context, and instantiates them into deterministic envelopes with set/version provenance. | | 📏 Tier B parameter bounds | Propagates explicit lower and upper parameter bounds through a deterministic scenario to produce min/max ranges, monotonicity checks, and bounded uncertainty records. | | 📈 Tier C probability bounds | Publishes packaged single-driver probability-bounds profiles with curated driver taxonomy and evaluates their support points without Monte Carlo or joint-distribution claims. | | 🧷 Tier C scenario packages | Publishes dependency-aware packaged scenario states with cumulative probability bounds, curated package taxonomy, preserved coupled drivers without Monte Carlo claims, and propagated mechanistic-constraint uncertainty notes when support-point scenarios activate bounded physics caps. | | 🚨 Tier 1 inhalation screening | Publishes machine-actionable upgrade advisories for spray inhalation scenarios, preserves the requestedTier routing hook on Tier 0 requests, ships a deterministic Tier 1 NF/FF screening tool, exposes packaged airflow, particle, and product-family screening profiles through a machine-readable manifest, warns when caller geometry or regime inputs diverge materially from matched profile anchors, and applies a bounded local-entrainment floor when very weak static interzonal mixing would otherwise become physically implausible. | | 🌫️ Residual-air reentry inhalation | Builds post-application room-air screening scenarios in two explicit modes: anchored reentry from a supplied start concentration, or native treated-surface reentry from bounded surface-emission plus room-loss terms, both with a low deposition sink and explicit non-application-plume semantics. | | 🔬 Evidence reconciliation and workflow | Normalizes CompTox, SCCS, SCCS opinions, CosIng, ConsExpo, nanomaterial guidance, microplastics regulatory records, and user-reviewed evidence into a shared product-use contract, ranks fit, builds merged requests, and can run an audited evidence-to-scenario-to-PBPK workflow in one response. | | 🧫 Particle-aware cosmetics and materials context | Publishes particle material context for EU cosmetic nanomaterials, synthetic polymer microparticles, and non-plastic micro/nanoparticles so route relevance, regulatory flags, and direct-use assumptions stay explicit without drifting into fate or toxicology claims. | | 🏭 Worker task routing | Routes worker-tagged tasks to the strongest current MCP path, emits worker-specific scenario guardrails when the shared screening engines are reused, and points higher-tier occupational cases toward future adapter hooks. | | 🔌 Worker Tier 2 bridge | Exports a typed worker inhalation handoff package, compatibility checklist, and future adapter tool-call envelope for ART-style Tier 2 refinement without pretending the occupational solver already exists. | | 🛠️ Worker Tier 2 execution | Executes a governed control-aware worker inhalation surrogate, supports bounded task-intensity inhalation-rate scaling plus explicit LEV/control-context, capture-zone, hood/enclosure-sensitive capture-distance decay, and LEV-family / hood-face-velocity refinements with bounded measured-profile bands, supports deterministic benchmark regression, and preserves comparability with external ART imports without claiming a native ART solver run. | | 🔄 Worker ART external exchange | Exports normalized external ART execution packages and imports reviewed external results or runner artifacts through a bounded, provenance-preserving adapter surface. | | 🧴 Worker dermal absorbed-dose execution | Exports and ingests dermal absorbed-dose/PPE handoffs, then executes a bounded dermal kernel with retained-loading/runoff caps, barrier-material and chemistry modifiers, bounded breakthrough-lag timing, duration-aware evaporation competition, and family-, subtype-, carrier-, formulation-, and physchem-aware pressurized-aerosol mass interpretation when only volumetric aerosol defaults are available, while keeping certified glove performance and full permeation modeling explicitly out of scope. | | 🧮 Aggregate summaries | Produces additive co-use summaries while preserving route and component transparency. | | 🧬 PBPK handoff export | Emits PBPK-ready objects plus an exact external-import package aligned to the upstream PBPK MCP request shape. | | 📤 Evidence export | Emits deterministic evidence, claim, and structured handoff primitives for downstream review and orchestration layers. | | 🔁 Refinement workflow support | Emits comparison/refinement bundles with explicit refine_exposure semantics and workflow hooks. | | ✅ Validation dossier | Publishes a typed validation dossier with benchmark domains, cited external validation datasets, heuristic-source families, and open evidence gaps, and threads evidence-readiness, executed validation checks, and gap IDs into every scenario-level validationSummary. | | 📋 Validation coverage report | Publishes a typed cross-domain trust summary over benchmark cases, external datasets, executable bands, time-series packs, and goldset mappings so validation posture is explicit instead of inferred. | | 🛡️ Verification summary | Publishes and executes a consolidated consistency check across release metadata, contract counts, benchmark coverage, validation assets, and published trust resources. | | 📐 Executable validation bands | Publishes a typed, versioned manifest for the narrow executable reference bands used by validationSummary.executedValidationChecks, so screening acceptance anchors are data-driven rather than hardcoded. | | ⏱️ Executable time-series packs | Publishes sparse governed time-series anchors for domains like residual-air reentry and air-space aerosol decay, so time-resolved validation is versioned and machine-readable. | | 🏅 Goldset showcase corpus | Publishes a separate, source-backed showcase set for recognizable cases while keeping the deterministic regression fixture stable and auditable. | | 🧷 Curated dermal contact packs | Replaces the highest-volume transfer and surface-contact-retention heuristics with RIVM-backed screening defaults for personal_care hand application and household_cleaner wipe contact while preserving explicit applicability domains and remaining evidence

…

Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: ToxMCP
• Source: ToxMCP/direct-use-exposure-mcp
• License: Apache-2.0

Install and usage instructions live in the source repository linked above.