肿瘤临床试验MCP服务器

 TRIALMCP -- END-TO-END ARCHITECTURE
 ==========================================

 +---------------------------+     +---------------------------+
 |    ROBOT AGENTS           |     |    HUMAN AGENTS           |
 |  (Physical AI Platforms)  |     |  (Trial Staff / Monitors) |
 |                           |     |                           |
 |  Franka Panda (USL 7.4)   |     |  Trial Coordinators       |
 |  da Vinci dVRK (USL 7.1)  |     |  Data Monitors            |
 |  Kinova Gen3  (USL 5.7)   |     |  Auditors                 |
 +-------------|-------------+     +-------------|-------------+
               |                                 |
               v                                 v
 +-----------------------------------------------------------+
 |              TRIALMCP-AUTHZ  (Policy Decision Point)      |
 |                                                           |
 |   RBAC Policy Engine        Token Management              |
 |   Deny-by-Default           Session Scoping               |
 |   Role: robot_agent         Role: trial_coordinator       |
 |   Role: data_monitor        Role: auditor                 |
 +-------------------------------|---------------------------+
                                 |
          +----------------------+----------------------+
          |                      |                      |
          v                      v                      v
 +----------------+   +------------------+   +-------------------+
 | TRIALMCP-FHIR  |   | TRIALMCP-DICOM   |   | TRIALMCP-LEDGER   |
 | (Read-Only)    |   | (Query/Retrieve) |   | (Audit Chain)     |
 |                |   |                  |   |                   |
 | Patient Lookup |   | C-FIND Proxy     |   | Hash-Chain Append |
 | Study Status   |   | C-MOVE Pointers  |   | Chain Verify      |
 | FHIR Search    |   | RECIST 1.1       |   | Replay Traces     |
 | De-ID Pipeline |   | Permission Gates |   | 21 CFR Part 11    |
 +----------|-----+   +---------|--------+   +---------|---------+
            |                   |                      |
            +-------------------+----------------------+
                                |
                                v
 +-----------------------------------------------------------+
 |           TRIALMCP-PROVENANCE  (Data Gateway)             |
 |                                                           |
 |   Source Registration      Access Recording               |
 |   Lineage Tracking         Integrity Verification         |
 |   Actor History            SHA-256 Fingerprinting         |
 |   Least-Privilege          Replayable Audit Traces        |
 +-----------------------------------------------------------+

%%{init: {'theme': 'base', 'themeVariables': {'primaryColor': '#e8f4f8', 'primaryBorderColor': '#5b9bd5', 'primaryTextColor': '#2c3e50', 'lineColor': '#5b9bd5', 'secondaryColor': '#fff3e0', 'tertiaryColor': '#e8f5e9', 'fontSize': '14px'}}}%%
flowchart LR
    subgraph BUILD ["v0.1.0 Build"]
        direction TB
        A["Server Code"]:::card
        B["Test Suite"]:::card
        C["Datasets"]:::card
    end
    subgraph REVIEW ["v0.1.1 Review"]
        direction TB
        D["@codex Audit"]:::reviewcard
        E["Recommendations"]:::reviewcard
        F["Gap Analysis"]:::reviewcard
    end
    subgraph HARDEN ["v0.2.0 Harden"]
        direction TB
        G["Validation"]:::greencard
        H["Error Codes"]:::greencard
        I["Token Expiry"]:::greencard
    end

    BUILD --> REVIEW --> HARDEN

    classDef card fill:#e8f4f8,stroke:#5b9bd5,stroke-width:2px,rx:12,ry:12,color:#2c3e50
    classDef reviewcard fill:#fff3e0,stroke:#e6a23c,stroke-width:2px,rx:12,ry:12,color:#2c3e50
    classDef greencard fill:#e8f5e9,stroke:#67c23a,stroke-width:2px,rx:12,ry:12,color:#2c3e50

 AUDIT HASH CHAIN -- 21 CFR PART 11 COMPLIANCE
 ===============================================

 +------------------+     +------------------+     +------------------+
 |  GENESIS BLOCK   |     |  AUDIT RECORD 1  |     |  AUDIT RECORD 2  |
 |                  |     |                  |     |                  |
 |  prev: 000...0   |---->|  prev: hash(G)   |---->|  prev: hash(R1)  |
 |  server: system  |     |  server: fhir    |     |  server: dicom   |
 |  tool: init      |     |  tool: fhir_read |     |  tool: dicom_qry |
 |  caller: system  |     |  caller: robot-1 |     |  caller: robot-1 |
 |  hash: SHA-256   |     |  hash: SHA-256   |     |  hash: SHA-256   |
 +------------------+     +------------------+     +------------------+
         |                                                    |
         |   CHAIN VERIFICATION: Recompute each hash and      |
         |   verify prev_hash links match. Any tampering,     |
         |   deletion, or reordering breaks the chain.        |
         +----------------------------------------------------+

 Properties:
   - Append-only: Records cannot be modified or deleted
   - Hash-linked: Each record includes hash of previous record
   - Tamper-evident: Any modification breaks the chain
   - Replayable: Full trace can be replayed for incident review
   - ICH-GCP E6(R2): Satisfies electronic records guidance

 TRIAL ROBOT AGENT -- WORKFLOW SEQUENCE
 ========================================

 Robot Agent                  TrialMCP Servers
 (Franka Panda)
      |
      |  1. AUTHENTICATE
      |--------------------->  trialmcp-authz
      |  Issue session token     |
      |<---------------------    | token_id + role
      |
      |  2. FETCH TASK ORDER
      |--------------------->  trialmcp-fhir
      |  Study status query      |
      |<---------------------    | study phase + enrollment
      |
      |  3. RETRIEVE IMAGING
      |--------------------->  trialmcp-dicom
      |  Get study pointer       |
      |<---------------------    | retrieval_token + endpoint
      |
      |  4. EXECUTE PROCEDURE
      |  (Robot performs task)
      |
      |  5. UPLOAD EVIDENCE
      |--------------------->  trialmcp-ledger
      |  Append audit record     |
      |<---------------------    | audit_id + record_hash
      |
      |  6. RECORD PROVENANCE
      |--------------------->  trialmcp-provenance
      |  Track data access       |
      |<---------------------    | record_id + confirmation
      |
      |  ALL STEPS AUDITED IN HASH CHAIN
      +

 FEDERATED MCP DEPLOYMENT -- MULTI-SITE TRIAL TOPOLOGY
 =======================================================

       SITE A                    SITE B                    SITE C
 +------------------+     +------------------+     +------------------+
 | trialmcp-fhir    |     | trialmcp-fhir    |     | trialmcp-fhir    |
 | trialmcp-dicom   |     | trialmcp-dicom   |     | trialmcp-dicom   |
 | trialmcp-authz   |     | trialmcp-authz   |     | trialmcp-authz   |
 | trialmcp-ledger  |     | trialmcp-ledger  |     | trialmcp-ledger  |
 +--------|---------+     +--------|---------+     +--------|---------+
          |                        |                        |
          |     Differential       |     Secure             |
          |     Privacy            |     Aggregation        |
          v                        v                        v
 +--------------------------------------------------------------------+
 |          FEDERATED COORDINATION LAYER                              |
 |                                                                    |
 |   FedAvg / FedProx / SCAFFOLD Aggregation                          |
 |   Cross-Site Audit Log Synchronization                             |
 |   DICOM/FHIR/LOINC/RxNorm/MedDRA Harmonization                     |
 |   Per-Site Privacy Budget Enforcement                              |
 +---------------------------------|----------------------------------+
                                   |
                                   v
 +--------------------------------------------------------------------+
 |           TRIALMCP-PROVENANCE  (Central Gateway)                   |
 |                                                                    |
 |   Cross-Site Lineage Tracking                                      |
 |   Data Fingerprint Verification                                    |
 |   Actor History Across Sites                                       |
 |   Replayable Multi-Site Audit Traces                               |
 +--------------------------------------------------------------------+

%%{init: {'theme': 'base', 'themeVariables': {'primaryColor': '#e8f4f8', 'primaryBorderColor': '#5b9bd5', 'primaryTextColor': '#2c3e50', 'lineColor': '#5b9bd5', 'fontSize': '14px'}}}%%
flowchart TB
    subgraph PHYS ["Physical AI Layer"]
        direction LR
        R1[("Robot\nPlatforms")]:::cyl
        SIM["Simulation\nFrameworks"]:::card
    end
    subgraph MCP ["MCP Protocol Layer"]
        direction LR
        AUTH["authz\nRBAC"]:::authcard
        FHIR["fhir\nClinical"]:::fhircard
        DICOM["dicom\nImaging"]:::dicomcard
        LED["ledger\nAudit"]:::ledcard
        PROV["provenance\nLineage"]:::provcard
    end
    subgraph CLIN ["Clinical Trial Layer"]
        direction LR
        EHR[("EHR\nSystems")]:::cyl
        PACS[("PACS\nArchives")]:::cyl
        REG["Regulatory\nCompliance"]:::regcard
    end

    PHYS --> MCP
    MCP --> CLIN
    AUTH --- FHIR
    AUTH --- DICOM
    FHIR --- LED
    DICOM --- LED
    LED --- PROV

    classDef card fill:#e8f4f8,stroke:#5b9bd5,stroke-width:2px,rx:12,ry:12,color:#2c3e50
    classDef cyl fill:#e8f4f8,stroke:#5b9bd5,stroke-width:2px,color:#2c3e50
    classDef authcard fill:#fde8e8,stroke:#f56c6c,stroke-width:2px,rx:12,ry:12,color:#2c3e50
    classDef fhircard fill:#e8f5e9,stroke:#67c23a,stroke-width:2px,rx:12,ry:12,color:#2c3e50
    classDef dicomcard fill:#f3e8ff,stroke:#9b59b6,stroke-width:2px,rx:12,ry:12,color:#2c3e50
    classDef ledcard fill:#fff3e0,stroke:#e6a23c,stroke-width:2px,rx:12,ry:12,color:#2c3e50
    classDef provcard fill:#e8f4f8,stroke:#409eff,stroke-width:2px,rx:12,ry:12,color:#2c3e50
    classDef regcard fill:#fde8e8,stroke:#f56c6c,stroke-width:2px,rx:12,ry:12,color:#2c3e50

 DATA FLOW SECURITY -- PRIVACY-PRESERVING ACCESS PATTERN
 =========================================================

 Agent Request              Security Layers                    Data Store
      |                          |                                |
      |  1. Token Presented      |                                |
      |------------------------->|                                |
      |                    [authz_evaluate]                       |
      |                    Role + Server + Tool                   |
      |                          |                                |
      |                    2. DENY by default                     |
      |                    Check ALLOW rules                      |
      |                    Check DENY overrides                   |
      |                          |                                |
      |                    3. Permission Granted                  |
      |                          |------------------------------->|
      |                          |     Raw clinical data          |
      |                          |<-------------------------------|
      |                          |                                |
      |                    4. De-identification                   |
      |                    HIPAA Safe Harbor (18 IDs)             |
      |                    HMAC-SHA256 pseudonymize               |
      |                          |                                |
      |                    5. Audit Record                        |
      |                    SHA-256 hash chain                     |
      |                    21 CFR Part 11                         |
      |                          |                                |
      |  6. De-identified data   |                                |
      |<-------------------------|                                |
      |                          |                                |

%%{init: {'theme': 'base', 'themeVariables': {'primaryColor': '#e8f4f8', 'primaryBorderColor': '#5b9bd5', 'primaryTextColor': '#2c3e50', 'lineColor': '#5b9bd5', 'fontSize': '14px'}}}%%
flowchart TB
    U["@kevinkawchak\nRequirements"]:::usercard
    CL["@claude\nImplementation"]:::claudecard
    CX["@codex\nPeer Review"]:::codexcard
    T["39 Tests\nAll Passing"]:::testcard
    V["Input Validation\nSSRF + UID"]:::valcard
    R["Publication\nReadiness"]:::pubcard

    U --> CL
    CL --> T
    T --> CX
    CX -->|recommendations| CL
    CL --> V
    V --> T
    T --> R

    classDef usercard fill:#e8f4f8,stroke:#5b9bd5,stroke-width:2px,rx:12,ry:12,color:#2c3e50
    classDef claudecard fill:#f3e8ff,stroke:#9b59b6,stroke-width:2px,rx:12,ry:12,color:#2c3e50
    classDef codexcard fill:#fff3e0,stroke:#e6a23c,stroke-width:2px,rx:12,ry:12,color:#2c3e50
    classDef testcard fill:#e8f5e9,stroke:#67c23a,stroke-width:2px,rx:12,ry:12,color:#2c3e50
    classDef valcard fill:#fde8e8,stroke:#f56c6c,stroke-width:2px,rx:12,ry:12,color:#2c3e50
    classDef pubcard fill:#e8f4f8,stroke:#409eff,stroke-width:2px,rx:12,ry:12,color:#2c3e50

git clone https://github.com/kevinkawchak/mcp-pai-oncology-trials.git
cd mcp-pai-oncology-trials
pip install pytest jsonschema

import json
from servers.trialmcp_fhir.src.fhir_server import FHIRMCPServer
from servers.trialmcp_dicom.src.dicom_server import DICOMMCPServer
from servers.trialmcp_ledger.src.ledger_server import LedgerMCPServer
from servers.trialmcp_authz.src.authz_server import AuthzMCPServer
from servers.trialmcp_provenance.src.provenance_server import ProvenanceMCPServer
from clients.reference_agent.src.trial_robot_agent import TrialRobotAgent

# Initialize servers
ledger = LedgerMCPServer()
fhir = FHIRMCPServer(audit_callback=lambda r: ledger.handle_tool_call(
    "ledger_append", {"server": r["server"], "tool": r["tool"],
     "caller_id": "system", "result_summary": r["result_summary"]}
))
dicom = DICOMMCPServer()
authz = AuthzMCPServer()
provenance = ProvenanceMCPServer()

# Load synthetic data
with open("datasets/fhir-bundles/oncology_trial_bundle.json") as f:
    fhir.store.load_bundle(json.load(f))
with open("datasets/dicom-samples/study_index.json") as f:
    for study in json.load(f)["studies"]:
        dicom.index.index_study(study)

# Run a robot agent workflow
agent = TrialRobotAgent(platform="Franka Panda")
result = agent.execute_sample_workflow(
    authz_server=authz,
    fhir_server=fhir,
    dicom_server=dicom,
    ledger_server=ledger,
    provenance_server=provenance,
    study_id="ONCO-TRIAL-2026-001",
    dicom_study_uid="1.2.826.0.1.3680043.8.1055.1.20260301.1",
)
print(f"Workflow completed: {result['workflow_id']}")

# Verify audit chain integrity
chain = ledger.handle_tool_call("ledger_verify", {})
print(f"Audit chain valid: {chain['valid']}, length: {chain['chain_length']}")

# Run all tests
python -m pytest tests/ -v

# Run security tests only
python -m pytest tests/security/ -v

# Run audit completeness tests
python -m pytest tests/audit/ -v

# Run integration tests
python -m pytest tests/integration/ -v

mcp-pai-oncology-trials/
├── servers/
│   ├── common/                 # Shared error codes, validation, health checks
│   │   └── __init__.py
│   ├── trialmcp_fhir/          # Read-only FHIR MCP server
│   │   └── src/
│   │       └── fhir_server.py
│   ├── trialmcp_dicom/         # DICOM query/retrieve MCP server
│   │   └── src/
│   │       └── dicom_server.py
│   ├── trialmcp_ledger/        # Audit ledger MCP server
│   │   └── src/
│   │       └── ledger_server.py
│   ├── trialmcp_authz/         # Authorization policy MCP server
│   │   └── src/
│   │       └── authz_server.py
│   └── trialmcp_provenance/    # Data provenance gateway
│       └── src/
│           └── provenance_server.py
├── clients/
│   └── reference_agent/        # Trial robot agent reference client
│       └── src/
│           └── trial_robot_agent.py
├── datasets/
│   ├── fhir-bundles/           # Synthetic FHIR R4 bundles
│   ├── dicom-samples/          # Synthetic DICOM study index
│   ├── scheduling/             # Simulated trial scheduling data
│   ├── manifest.json           # SHA-256 checksums for reproducibility
│   └── README.md               # Data dictionary and provenance
├── tests/
│   ├── security/               # SSRF, injection, escalation, replay, health tests
│   ├── audit/                  # Audit completeness tests
│   └── integration/            # End-to-end robot workflow tests
├── peer-review/                # Peer-review recommendations and responses
├── pyproject.toml              # Project configuration
├── changelog.md                # Version changelog
├── releases.md                 # Release notes
├── prompts.md                  # Build prompts
├── SECURITY.md                 # Security policy
├── CONTRIBUTING.md             # Contribution guidelines
├── CITATION.cff                # Academic citation metadata
├── LICENSE                     # MIT License
└── README.md                   # This file

 TOOL SURFACE AREA -- 23 MCP TOOLS ACROSS 5 SERVERS
 =====================================================

 trialmcp-fhir (4 tools)          trialmcp-dicom (4 tools)
 +--------------------------+     +--------------------------+
 | fhir_read                |     | dicom_query              |
 | fhir_search              |     | dicom_retrieve_pointer   |
 | fhir_patient_lookup      |     | dicom_study_metadata     |
 | fhir_study_status        |     | dicom_recist_measurements|
 +--------------------------+     +--------------------------+

 trialmcp-ledger (5 tools)         trialmcp-authz (5 tools)
 +--------------------------+     +--------------------------+
 | ledger_append            |     | authz_evaluate           |
 | ledger_verify            |     | authz_issue_token        |
 | ledger_query             |     | authz_validate_token     |
 | ledger_replay            |     | authz_list_policies      |
 | ledger_chain_status      |     | authz_revoke_token       |
 +--------------------------+     +--------------------------+

 trialmcp-provenance (5 tools)
 +--------------------------+
 | provenance_register_src  |
 | provenance_record_access |
 | provenance_get_lineage   |
 | provenance_get_actor_hst |
 | provenance_verify_intgty |
 +--------------------------+