biocTcgaMcp
Purpose: demo an MCP server with tools for TCGA analhsis
With this package installed and a suitable configuration file at ~/.config/mcptools/config.jsonthat includes
{
"mcpServers": {
"r-mcptools": {
"command": "Rscript",
"args": ["-e", "mcptools::mcp_server()"]
},
"r-btw": {
"command": "Rscript",
"args": ["-e", "btw::btw_mcp_server()"]
},
"r-tcga": {
"command": "Rscript",
"args": ["-e", "biocTcgaMcp::btcga_mcp_server()"]
}
}
}
then
library(ellmer)
ch = chat_openai(model="gpt-4o")
ch$set_tools(mcptools::mcp_tools()) # need biocTcgaMcp 0.0.2 for improved type list tool
cc = type_array(type_object(codes = type_string(description = "TCGA tumor type codes"),
types = type_string(description = "TCGA tumor study names"))
ch$chat_structured("What tumor types are available in TCGA?", type=cc)
should return a data.frame. chat_structured is not necessary but createsa value in the calling R session.
> ch$chat_structured("What tumor types are available in TCGA?", type=cc)
codes types
1 ACC Adrenocortical carcinoma
2 BLCA Bladder Urothelial Carcinoma
3 BRCA Breast invasive carcinoma
4 CESC Cervical squamous cell carcinoma and endocervical adenocarcinoma
5 CHOL Cholangiocarcinoma
6 COAD Colon adenocarcinoma
7 DLBC Lymphoid Neoplasm Diffuse Large B-cell Lymphoma
8 ESCA Esophageal carcinoma
9 GBM Glioblastoma multiforme
10 HNSC Head and Neck squamous cell carcinoma
11 KICH Kidney Chromophobe
12 KIRC Kidney renal clear cell carcinoma
13 KIRP Kidney renal papillary cell carcinoma
14 LAML Acute Myeloid Leukemia
15 LGG Brain Lower Grade Glioma
16 LIHC Liver hepatocellular carcinoma
17 LUAD Lung adenocarcinoma
18 LUSC Lung squamous cell carcinoma
19 MESO Mesothelioma
20 OV Ovarian serous cystadenocarcinoma
21 PAAD Pancreatic adenocarcinoma
22 PCPG Pheochromocytoma and Paraganglioma
23 PRAD Prostate adenocarcinoma
24 READ Rectum adenocarcinoma
25 SARC Sarcoma
26 SKCM Skin Cutaneous Melanoma
27 STAD Stomach adenocarcinoma
28 TGCT Testicular Germ Cell Tumors
29 THCA Thyroid carcinoma
30 THYM Thymoma
31 UCEC Uterine Corpus Endometrial Carcinoma
32 UCS Uterine Carcinosarcoma
33 UVM Uveal Melanoma
The tool defined in this package that should be used to answer the query above canbe run directly with biocTcgaMcp::list_cancer_types() and returns a data.frame with33 rows.
A non-MCP query result:
> ch = chat_openai(model="gpt-4o")
> ch$chat("What are the tumor types studied in TCGA?")
The Cancer Genome Atlas (TCGA) was a large-scale project aimed at cataloging genetic mutations responsible for
different types of cancer using genome sequencing and bioinformatics. TCGA studied a wide variety of tumor types,
including but not limited to:
1. **Breast Invasive Carcinoma (BRCA)**
2. **Lung Adenocarcinoma (LUAD)**
3. **Lung Squamous Cell Carcinoma (LUSC)**
4. **Prostate Adenocarcinoma (PRAD)**
5. **Kidney Renal Clear Cell Carcinoma (KIRC)**
6. **Thyroid Carcinoma (THCA)**
7. **Stomach Adenocarcinoma (STAD)**
8. **Liver Hepatocellular Carcinoma (LIHC)**
9. **Colorectal Adenocarcinoma (COADREAD)**
10. **Brain Lower Grade Glioma (LGG)**
11. **Glioblastoma Multiforme (GBM)**
12. **Ovarian Serous Cystadenocarcinoma (OV)**
13. **Pancreatic Adenocarcinoma (PAAD)**
14. **Bladder Urothelial Carcinoma (BLCA)**
15. **Head and Neck Squamous Cell Carcinoma (HNSC)**
16. **Uterine Corpus Endometrial Carcinoma (UCEC)**
17. **Skin Cutaneous Melanoma (SKCM)**
18. **Esophageal Carcinoma (ESCA)**
19. **Cervical Squamous Cell Carcinoma and Endocervical Adenocarcinoma (CESC)**
20. **Acute Myeloid Leukemia (LAML)**
21. **Kidney Renal Papillary Cell Carcinoma (KIRP)**
22. **Kidney Chromophobe (KICH)**
23. **Uveal Melanoma (UVM)**
24. **Mesothelioma (MESO)**
25. **Pheochromocytoma and Paraganglioma (PCPG)**
26. **Thymoma (THYM)**
27. **Adrenocortical Carcinoma (ACC)**
28. **Cholangiocarcinoma (CHOL)**
These tumor types were selected for their prevalence and mortality rates, as well as their scientific and clinical
importance. The data collected from TCGA has been crucial in advancing the understanding of the genomic underpinnings
of cancer.
That's not too satisfying because we know there are 33 tumor types.
The purpose of this package is to refine tool definition and prompt design so that wehave low risk of getting an error.
