"""
Model Context Protocol (MCP) for Paper2Agent

This codebase provides comprehensive bioinformatics analysis tools for transcriptomics data using the decoupler package. It supports bulk RNA-seq, single-cell RNA-seq, spatial transcriptomics, and pseudotime analysis workflows. The tools enable enrichment analysis, transcription factor activity scoring, pathway analysis, and cross-species gene mapping for functional genomics research.

DATASET DECISION TREE
═══════════════════════════════════════════════════════════════════════
STEP 0 (if data lives in GEO):
  → decoupler_load_geo_series_matrix(url_or_path, condition_column=...)
  This downloads, parses, and saves an h5ad. Then proceed to STEP 1.

STEP 1 — Always start with: decoupler_inspect_data(adata_path)
  Check: is_integer, has_negatives, likely_raw_counts, n_obs, orientation_warning

STEP 2 — If orientation_warning is set:
  The matrix is transposed (genes as rows). Fix it before proceeding.

STEP 3 — Choose path based on inspect results:

  PATH A — Raw counts (likely_raw_counts=True)
    decoupler_load_and_filter_data
    → decoupler_preprocess_data
    → decoupler_differential_expression(method='deseq2')
    → enrichment tools

  PATH B — Pre-normalized / log-transformed (data_type='log_expression' or 'log_ratio_microarray')
    Examples: GEO microarray, log-CPM, log-TPM, log-normalized scRNA-seq
    SKIP decoupler_preprocess_data entirely.
    Step 2a (microarray only — if features_look_like_probes=True):
      → decoupler_collapse_probes_to_genes  # map probe IDs → gene symbols before DE
      Skip this step for RNA-seq (features are already gene symbols).
    Step 2b:
      → decoupler_differential_expression(method='ttest')   # no R required
      → decoupler_differential_expression(method='limma')   # preferred if R available
    → enrichment tools
    ⚠ Do NOT use method='deseq2' or decoupler_preprocess_data on pre-normalized data.

  PATH UNKNOWN — data_type='unknown' (heuristics inconclusive)
    Default to PATH B with method='ttest'. Ask the user to confirm data is not raw
    integer counts before proceeding. If it is, switch to PATH A with method='deseq2'.
═══════════════════════════════════════════════════════════════════════

This MCP Server contains tools extracted from the following tutorial files:
1. example
    - decoupler_load_toy_data: Load toy dataset and visualize gene expression patterns
    - decoupler_run_individual_methods: Run individual enrichment methods (GSEA and ULM)
    - decoupler_run_multiple_methods: Run all enrichment methods and compute consensus scores
    - decoupler_access_prior_knowledge: Query OmniPath resources for prior knowledge networks
2. licenses
    - decoupler_collect_transcriptional_regulators: Collect transcriptional regulator data with specified license
    - decoupler_access_resource_database: Access specific OmniPath resources with license restrictions
3. orthologs
    - decoupler_show_supported_organisms: Display organisms supported for ortholog translation
    - decoupler_load_knowledge_base: Load human-centric knowledge database resource
    - decoupler_translate_gene_symbols: Convert gene symbols from human to target organism
    - decoupler_access_organism_database: Directly load database with organism-specific gene symbols
4. rna
    - decoupler_inspect_data: Inspect data structure, detect orientation and data type, recommend analysis path
    - decoupler_load_geo_series_matrix: Download and parse a GEO series matrix URL or file into h5ad
    - decoupler_annotate_probes_with_gpl: Add gene symbols to var from a GPL platform SOFT file (required before probe collapse)
    - decoupler_collapse_probes_to_genes: Collapse microarray probe IDs to gene symbols before enrichment
    - decoupler_load_and_filter_data: Load bulk RNA-seq data and apply expression filtering
    - decoupler_preprocess_data: Normalize, scale and perform PCA analysis on bulk RNA-seq data (raw counts only)
    - decoupler_differential_expression: Differential expression via DESeq2, limma, or t-test (method parameter)
    - decoupler_tf_enrichment_collectri: TF enrichment using CollecTRI; auto-selects top TFs for network plot
    - decoupler_pathway_enrichment_progeny: Perform pathway enrichment analysis using PROGENy
    - decoupler_hallmark_enrichment: Perform hallmark gene set enrichment analysis
5. rna_pstime
    - decoupler_load_and_explore_data: Load and explore single-cell data with initial visualizations
    - decoupler_infer_pseudotime: Infer pseudotime trajectories using diffusion maps and DPT
    - decoupler_analyze_tf_activity: Analyze transcription factor activity along pseudotime
    - decoupler_analyze_pathway_activity: Analyze pathway activity using PROGENy along pseudotime
    - decoupler_analyze_hallmark_pathways: Analyze Hallmark pathway activity along pseudotime
6. rna_sc
    - decoupler_load_and_visualize_data: Load single-cell data and create initial UMAP visualization
    - decoupler_score_cell_types: Score cells using marker genes from PanglaoDB
    - decoupler_score_transcription_factors: Score cells using transcription factor networks from CollecTRI
    - decoupler_score_progeny_pathways: Score cells using PROGENy pathway gene sets
    - decoupler_score_hallmark_pathways: Score cells using Hallmark gene sets
7. rna_visium
    - decoupler_load_and_preprocess_visium: Load and preprocess spatial transcriptomics data with normalization
    - decoupler_apply_spatial_weighting: Apply spatial connectivity weighting to gene expression data
    - decoupler_score_transcription_factors: Score transcription factor activity using CollecTRI network
    - decoupler_score_progeny_pathways: Score pathway activity using PROGENy pathway database
    - decoupler_score_hallmark_pathways: Score pathway activity using Hallmark gene sets
"""

import os
import sys
import warnings

PROJECT_ROOT = os.path.dirname(os.path.abspath(__file__))
if PROJECT_ROOT not in sys.path:
    sys.path.insert(0, PROJECT_ROOT)

# ── R dependency installation ────────────────────────────────────────────────
# Run once at startup so limma is available when decoupler_differential_expression
# is called with method='limma'.  Failures are non-fatal: the rna.py tool
# handles a missing limma gracefully via fallback_to_ttest or RuntimeError.
try:
    from src.mcp_server.dependencies.r_dependencies import ensure_r_dependencies

    ensure_r_dependencies()
except FileNotFoundError as _e:
    warnings.warn(f"R install script not found, skipping R setup: {_e}", stacklevel=2)
except Exception as _e:
    warnings.warn(f"R dependency installation failed, limma may be unavailable: {_e}", stacklevel=2)

from fastmcp import FastMCP

# Import statements (alphabetical order)
from src.tools.bulk_dataset_tools import bulk_dataset_mcp
from src.tools.bulk_rnaseq import bulk_rnaseq_mcp
from src.tools.dataset_tools import dataset_mcp
from src.tools.example import example_mcp
from src.tools.integration_tools import integration_mcp
from src.tools.licenses import licenses_mcp
from src.tools.orthologs import orthologs_mcp
from src.tools.rna import rna_mcp
from src.tools.rna_pstime import rna_pstime_mcp
from src.tools.rna_sc import rna_sc_mcp
from src.tools.rna_visium import rna_visium_mcp

# Server definition and mounting
mcp = FastMCP(name="Paper2Agent")
mcp.mount(bulk_dataset_mcp)
mcp.mount(bulk_rnaseq_mcp)
mcp.mount(dataset_mcp)
mcp.mount(example_mcp)
mcp.mount(integration_mcp)
mcp.mount(licenses_mcp)
mcp.mount(orthologs_mcp)
mcp.mount(rna_mcp)
mcp.mount(rna_pstime_mcp)
mcp.mount(rna_sc_mcp)
mcp.mount(rna_visium_mcp)

import threading

from src.cache import preload_datasets

# Pre-populate the disk cache in a background thread (pure Python/numpy/scanpy,
# no rpy2 — safe to run concurrently with tool calls).
# R/limma installation is handled at module import above (already synchronous).
threading.Thread(target=preload_datasets, daemon=True, name="cache-preloader").start()

if __name__ == "__main__":
    import argparse

    parser = argparse.ArgumentParser(add_help=False)
    parser.add_argument("--transport", default=None)
    parser.add_argument("--port", type=int, default=8765)
    args, _ = parser.parse_known_args()

    if args.transport:
        # HTTP mode: persistent server, accepts multiple connections
        mcp.run(transport=args.transport, host="0.0.0.0", port=args.port)
    else:
        # stdio mode: default for MCP clients that spawn a subprocess per call
        mcp.run()