Advanced Claude Code Tools Setup

A comprehensive guide to advanced Claude Code features and a full "BioClaude" setup for biological research. Covers hooks, skills, custom agents, MCP servers for life sciences, and future-proofing strategies.

Who this is for: Researchers (especially in biology/genomics) who want to turn Claude Code into a full-powered research workstation connected to biological databases, bioinformatics tools, and lab platforms.

Author: Robert Chen (Fowler Lab, UW) Last updated: 2026-02-16 (install instructions verified against actual repos)

Advanced Claude Code Features
MCP (Model Context Protocol) Overview
BioClaude: Full Bio Research Setup
MCP Server Reference Table
Future-Proofing: Stay Up to Date
Recommended Project Structure
Troubleshooting

1. Advanced Claude Code Features

These are built-in Claude Code capabilities that don't require MCP. Set them up by creating a .claude/ directory in your project root.

mkdir -p .claude/skills .claude/agents

Hooks

Hooks are shell commands that fire automatically at lifecycle events. They live in .claude/settings.json.

Available hook events:

Event	When It Fires	Use Case
`SessionStart`	When a session begins	Print reminders, load environment
`PreToolUse`	Before a tool executes	Block dangerous operations, enforce rules
`PostToolUse`	After a tool succeeds	Auto-run tests, lint, format
`Stop`	When Claude finishes responding	Summary logging, notifications

Example: Auto-run tests after every file edit

// .claude/settings.json
{
  "hooks": {
    "PostToolUse": [
      {
        "matcher": "Edit|Write",
        "hooks": [
          {
            "type": "command",
            "command": "cd $CLAUDE_PROJECT_DIR && python -m pytest test_calculator.py -q 2>&1 | tail -5"
          }
        ]
      }
    ]
  }
}

Example: Block edits to production config files

{
  "hooks": {
    "PreToolUse": [
      {
        "matcher": "Edit|Write",
        "hooks": [
          {
            "type": "command",
            "command": "echo $CLAUDE_TOOL_INPUT | grep -q 'production.yml' && exit 2 || exit 0"
          }
        ]
      }
    ]
  }
}

Exit code 2 from a PreToolUse hook blocks the operation. Exit code 0 allows it.

Example: Session start reminder

{
  "hooks": {
    "SessionStart": [
      {
        "hooks": [
          {
            "type": "command",
            "command": "echo '** Remember: run tests before committing. Check CLAUDE.md for project rules. **'"
          }
        ]
      }
    ]
  }
}

Skills (Custom Slash Commands)

Skills are reusable prompt templates you invoke with /skill-name. They live in .claude/skills/<name>/SKILL.md.

Creating a skill:

mkdir -p .claude/skills/fix-bug

Then create .claude/skills/fix-bug/SKILL.md:

---
name: fix-bug
description: Fix a bug in the calculator with full test coverage
argument-hint: "[function-name]"
---

Fix the bug in $ARGUMENTS in calculator.py:
1. Write a failing test that reproduces the bug
2. Fix the implementation
3. Run the full test suite to verify nothing broke
4. Commit with message: "fix: handle edge case in [function-name]"

Usage: /fix-bug divide

Dynamic context injection — inject shell command output into the prompt:

---
name: review
description: Review recent changes with test coverage info
---

Review the following changes and suggest improvements:

!`git diff HEAD~1`!

Current test coverage:
!`python -m pytest --cov=calculator --cov-report=term-missing -q 2>&1 | tail -20`!

More skill ideas for a bio research project:

Skill	What It Does
`/run-tests`	Run pytest with coverage report
`/lint`	Run linting and type checking
`/search-lit [topic]`	Search PubMed via MCP for a topic
`/lookup-protein [id]`	Query UniProt + AlphaFold for a protein
`/enrichment [gene-list]`	Run Enrichr enrichment analysis

Custom Agents (Subagents)

Custom agents are specialized Claude instances with their own tools, models, and instructions. They live in .claude/agents/<name>.md.

Example: Test runner agent

Create .claude/agents/test-runner.md:

---
name: test-runner
description: Run tests and diagnose failures
tools: Bash, Read
model: haiku
---

Run the full test suite. For any failures:
1. Read the failing test and the function under test
2. Identify the root cause
3. Suggest a specific fix with code
4. Do NOT modify any files — only report findings

Example: Literature search agent

Create .claude/agents/lit-search.md:

---
name: lit-search
description: Search biological literature across PubMed and preprint servers
tools: Bash, Read, WebSearch, WebFetch
model: sonnet
---

Search for papers related to the user's query. For each relevant paper:
1. Provide title, authors, journal, year
2. Write a 2-3 sentence summary of the key findings
3. Note any methodological relevance to deep mutational scanning or functional genomics
4. Provide the DOI or URL

Search PubMed, bioRxiv, and medRxiv. Prioritize recent papers (last 2 years).

Invoking agents:

Use /agents to list available agents
Agents are invoked automatically when Claude determines they're relevant, or you can reference them explicitly in your prompts

Extended Thinking

For complex analysis, you can trigger deeper reasoning by phrasing your prompts:

Phrase	Effect
"think step by step"	Standard chain-of-thought
"think deeply about this"	Extended thinking mode
"ultrathink"	Maximum reasoning depth

Useful for: debugging complex pipelines, analyzing experimental designs, planning multi-step bioinformatics workflows.

Permissions Whitelisting

Reduce permission prompts for tools you trust by configuring allowed tools in .claude/settings.json:

{
  "permissions": {
    "allow": [
      "Bash(pytest:*)",
      "Bash(python:*)",
      "Bash(git:*)",
      "Bash(conda:*)",
      "Bash(pip:*)"
    ]
  }
}

This lets Claude run pytest, python, git, conda, and pip commands without asking for permission each time.

2. MCP (Model Context Protocol) Overview

MCP is an open protocol that connects Claude to external tools and data sources. Think of it like USB-C for AI — a standard way to plug in any tool.

How it works:

Claude Code  <-->  MCP Client  <-->  MCP Server  <-->  External Tool/Database

Two transport types:

Type	How It Works	When to Use
`http` (remote)	Connects to a hosted server via URL	Cloud-hosted databases (UniProt, PubMed, etc.)
`stdio` (local)	Runs a local process that Claude communicates with via stdin/stdout	CLI tools (BLAST, SAMtools), local APIs

Adding an MCP server:

# Remote (HTTP) server
claude mcp add --transport http <name> <url>

# Local (stdio) server
claude mcp add --transport stdio <name> <command> [args...]

# List configured servers
claude mcp list

# Remove a server
claude mcp remove <name>

MCP configuration is stored in .claude/settings.json under the mcpServers key.

3. BioClaude: Full Bio Research Setup

A step-by-step guide to building a Claude Code environment optimized for biological research — from protein databases to bioinformatics pipelines.

Quick Start (5 minutes)

Get the most value with the least setup. These three commands give you access to 20+ biological databases, scientific figure search, and literature search.

# Step 1: BioContextAI — unified access to UniProt, AlphaFold, Ensembl,
# KEGG, Gene Ontology, Europe PMC, ClinicalTrials.gov, and more
# Published in Nature Biotechnology (2025). One server, 20+ databases.
claude mcp add --transport http biocontext https://mcp.biocontext.ai/mcp/

# Step 2: BioRender — search 50,000+ scientifically accurate icons and templates
# Official Anthropic partnership. Requires free BioRender account.
claude mcp add --transport http biorender https://mcp.services.biorender.com/mcp

# Step 3: BioMCP — PubMed + ClinicalTrials.gov + genomic variants (21 tools)
pip install biomcp-python
claude mcp add --transport stdio biomcp -- biomcp run

# Verify everything is connected
claude mcp list

Test it out:

> Search UniProt for human BRCA1 and show me the protein domains

> Find recent papers on deep mutational scanning of BRCA1

> Search BioRender for a CRISPR illustration

That's it. You now have a research-connected Claude. The sections below go deeper.

Official Anthropic Life Sciences Connectors

Anthropic maintains an official set of life sciences connectors through the anthropics/life-sciences GitHub marketplace. These are the connectors featured on the Claude for Life Sciences page.

Claude Desktop (claude.ai): Most of these appear in Settings > Connectors (for remote HTTP servers) or Settings > Extensions (for local bundles). Just search by name and enable.

Claude Code (CLI): Use claude mcp add as shown below.

Scientific Research & Discovery

# PubMed — millions of biomedical research articles and clinical studies, no auth required
# Official Anthropic-hosted connector.
# Tutorial: https://claude.com/resources/tutorials/using-the-pubmed-connector-in-claude
claude mcp add --transport http pubmed https://pubmed.mcp.claude.com/mcp

# bioRxiv / medRxiv — 260,000+ preprints, no auth required
# By deepsense.ai. Docs: https://docs.mcp.deepsense.ai/guides/biorxiv.html
claude mcp add --transport http biorxiv https://mcp.deepsense.ai/biorxiv/mcp

# ChEMBL — 2.4M+ bioactive compounds, targets, activity measurements, no auth required
# By deepsense.ai. Docs: https://docs.mcp.deepsense.ai/guides/chembl.html
claude mcp add --transport http chembl https://mcp.deepsense.ai/chembl/mcp

# Open Targets — drug target identification and prioritization (via BioContextAI above)
# Already covered by the BioContextAI server in Quick Start.

# Synapse.org — discover biomedical datasets, project structure, data assets
# By Sage Bionetworks. Requires free Synapse account (https://www.synapse.org).
# Auth: OAuth2 browser login on first use.
# GitHub: https://github.com/Sage-Bionetworks/synapse-mcp
claude mcp add --transport http synapse https://mcp.synapse.org/mcp

# Scholar Gateway by Wiley — authenticated access to scientific research papers
# Requires free CONNECT account (Wiley SSO). Trial access through Feb 28, 2026.
# Docs: https://docs.scholargateway.ai/
claude mcp add --transport http scholar-gateway https://connector.scholargateway.ai/mcp

# Owkin Pathology Explorer — analyze H&E pathology slides (TCGA), quantify cell
# types, spatial tumor microenvironment analysis, cohort-level survival analysis
# Auth required (Owkin account).
claude mcp add --transport http owkin https://mcp.k.owkin.com/mcp

Clinical Operations & Regulatory

# ClinicalTrials.gov — 500,000+ clinical studies from NIH/NLM registry, no auth required
# By deepsense.ai. Docs: https://docs.mcp.deepsense.ai/guides/clinical_trials.html
claude mcp add --transport http clinical-trials https://mcp.deepsense.ai/clinical_trials/mcp

# Medidata — clinical trial platform data (Rave EDC, predictive site ranking)
# Requires Medidata platform credentials. Site Ranking needs Intelligent Trials subscription.
# Tutorial: https://claude.com/resources/tutorials/using-the-medidata-connector-in-claude
claude mcp add --transport http medidata https://mcp.imedidata.com/mcp

Local Extensions (Desktop/stdio)

These run locally on your machine rather than connecting to a remote server.

# 10x Genomics Cloud — single cell and spatial analysis
# Requires Node.js and a 10x Genomics Cloud access token.
# Generate token at: https://cloud.10xgenomics.com/account/security
# GitHub: https://github.com/10XGenomics/txg-mcp
# Docs: https://www.10xgenomics.com/support/software/cloud-analysis/latest/tutorials/cloud-mcp-server-code
#
# For Claude Desktop: Settings > Extensions > "10x Genomics" > Install
# For Claude Code: follow the plugin method below or see the 10x docs for manual setup.

# ToolUniverse — 600+ vetted scientific tools (ML models, datasets, APIs)
# By MIMS Harvard (Marinka Zitnik Lab).
# GitHub: https://github.com/mims-harvard/ToolUniverse
# Docs: https://zitniklab.hms.harvard.edu/ToolUniverse/guide/building_ai_scientists/claude_code.html
pip install tooluniverse
claude mcp add --transport stdio tooluniverse -- tooluniverse-smcp-stdio --compact-mode

# Optional: load only specific tool categories
# claude mcp add --transport stdio tooluniverse -- tooluniverse-smcp-stdio --compact-mode --tool-categories uniprot,chembl,opentarget

# Optional: set API keys for enhanced access
# export NCBI_API_KEY=your_key
# export SEMANTIC_SCHOLAR_API_KEY=your_key

Tier 1: Core Bio Databases

These MCP servers connect Claude directly to the major biological databases.

BioContextAI (Recommended Starting Point)

A unified MCP gateway to 20+ biomedical databases. Published in Nature Biotechnology by researchers at EMBL-EBI, Helmholtz Munich, and Open Targets.

# Remote hosted server (for testing and moderate use — subject to fair use)
claude mcp add --transport http biocontext https://mcp.biocontext.ai/mcp/

# OR: Run locally for heavier use
pip install biocontext-kb
claude mcp add --transport stdio biocontext-local -- uvx biocontext_kb@latest

Databases accessible through BioContextAI:

Database	What It Provides
UniProt	Protein sequences, functions, annotations
AlphaFold DB	Predicted 3D protein structures
Ensembl	Gene sequences, coordinates, variants
KEGG	Metabolic and signaling pathways
Gene Ontology	Functional annotations (GO terms)
Europe PMC	Biomedical literature (like PubMed but broader)
ClinicalTrials.gov	Clinical trial records
ChEMBL	Bioactive drug-like molecules
Open Targets	Drug target evidence and associations
Reactome	Curated biological pathways

GitHub: biocontext-ai/knowledgebase-mcp
Registry: biocontext.ai/registry

Individual Database Servers (Augmented Nature Suite)

If you need deeper access to specific databases beyond what BioContextAI provides.

Prerequisites: Node.js (v16+) and npm. All four servers are TypeScript projects.

# UniProt — 26 tools: protein search, sequence retrieval, feature analysis, homologs
# GitHub: Augmented-Nature/Augmented-Nature-UniProt-MCP-Server
git clone https://github.com/Augmented-Nature/Augmented-Nature-UniProt-MCP-Server.git
cd Augmented-Nature-UniProt-MCP-Server && npm install && npm run build
claude mcp add --transport stdio uniprot -- node build/index.js

# AlphaFold — structure retrieval, confidence analysis, multi-format downloads
# GitHub: Augmented-Nature/AlphaFold-MCP-Server
git clone https://github.com/Augmented-Nature/AlphaFold-MCP-Server.git
cd AlphaFold-MCP-Server && npm install && npm run build
claude mcp add --transport stdio alphafold -- node build/index.js

# PDB — experimental structures, quality metrics, cross-references
# GitHub: Augmented-Nature/PDB-MCP-Server
git clone https://github.com/Augmented-Nature/PDB-MCP-Server.git
cd PDB-MCP-Server && npm install && npm run build
claude mcp add --transport stdio pdb -- node build/index.js

# Reactome — pathway search, gene-to-pathways, disease pathways (8 tools)
# GitHub: Augmented-Nature/Reactome-MCP-Server
git clone https://github.com/Augmented-Nature/Reactome-MCP-Server.git
cd Reactome-MCP-Server && npm install && npm run build
claude mcp add --transport stdio reactome -- node build/index.js

Note: For the claude mcp add commands above, use the absolute path to build/index.js (e.g., node /home/user/Augmented-Nature-UniProt-MCP-Server/build/index.js).

Tier 2: Lab Platforms

BioRender

Official Anthropic partnership. Search 50,000+ scientifically accurate icons and templates for figures, posters, graphical abstracts.

claude mcp add --transport http biorender https://mcp.services.biorender.com/mcp

Requires a free BioRender account (premium features need paid plan)
BioRender MCP Docs

Benchling

Official MCP server for querying your lab's ELN (electronic lab notebook), sample registry, and inventory.

# Replace <tenant> with your Benchling tenant name
claude mcp add --transport http benchling https://<tenant>.mcp.benchling.com/mcp

Prerequisites:

Your Benchling admin must enable the V3 API and MCP access
Benchling AI (Deep Research) module must be active on your tenant
Queries inherit your Benchling permissions and are fully audited

What you can do:

Search across experimental results and notebook entries
Query the sample registry and inventory
Find protocols and templates
Cross-reference data across experiments
Benchling MCP Docs
Community alternative: pip install benchling-mcp (unofficial)

Protocols.io

Search, retrieve, and manage scientific protocols.

# GitHub: hqn21/protocols-io-mcp-server (Python project, not Node)
pip install protocols-io-mcp
claude mcp add --transport stdio protocols -- protocols-io-mcp

Tier 3: Bioinformatics CLI Tools

The bio-mcp project wraps standard bioinformatics command-line tools as MCP servers. This lets you say things like "BLAST this sequence against nr" and Claude orchestrates the entire pipeline.

Prerequisites: Install the underlying tools first.

# Install bioinformatics tools via conda/mamba
conda install -c bioconda blast samtools bwa seqkit fastqc multiqc

Available bio-mcp servers:

Server	Tool	What It Does
`bio-mcp-blast`	NCBI BLAST	Sequence similarity search, custom DB creation
`bio-mcp-samtools`	SAMtools	BAM/SAM manipulation, sorting, indexing, flagstat
`bio-mcp-bwa`	BWA	Read alignment to reference genomes
`bio-mcp-seqkit`	SeqKit	Sequence stats, format conversion, filtering
`bio-mcp-fastqc`	FastQC/MultiQC	Sequencing data quality control

# Example: Install and configure bio-mcp-blast (not on PyPI — install from source)
git clone https://github.com/bio-mcp/bio-mcp-blast.git
cd bio-mcp-blast && pip install -e .
claude mcp add --transport stdio blast -- bio-mcp-blast

# Example: Install and configure bio-mcp-samtools (not on PyPI — install from source)
git clone https://github.com/bio-mcp/bio-mcp-samtools.git
cd bio-mcp-samtools && pip install -e .
claude mcp add --transport stdio samtools -- bio-mcp-samtools

Example conversation after setup:

> I have a FASTA file at data/unknown_sequences.fasta.
> BLAST these against the nr database and summarize the top hits.

> Align the FASTQ files in data/reads/ to the hg38 reference genome,
> sort the output, and give me alignment statistics.

Tier 4: Literature & Enrichment

PubMed MCP Server

# Option A: Dedicated PubMed server (cyanheads) — requires Node.js >= 20
# GitHub: cyanheads/pubmed-mcp-server
# Easiest: use npx (no manual clone needed)
claude mcp add --transport stdio pubmed -- npx @cyanheads/pubmed-mcp-server

# Option B: Multi-source paper search (arXiv + PubMed + bioRxiv + medRxiv)
# GitHub: openags/paper-search-mcp
pip install paper-search-mcp
claude mcp add --transport stdio papers -- paper-search-mcp

BioMCP (GenomOncology)

A comprehensive server combining literature, clinical trials, and genomic variants.

pip install biomcp-python
claude mcp add --transport stdio biomcp -- biomcp run

21 tools including:

PubMed article search and retrieval
ClinicalTrials.gov query
MyVariant.info genomic variant lookup
Cross-referencing between all three
GitHub: genomoncology/biomcp

Enrichr MCP Server

Gene set enrichment analysis across GO, KEGG, Reactome, WikiPathways, and more.

# GitHub: tianqitang1/enrichr-mcp-server (TypeScript project, not Python)
# Easiest: use npx (no manual clone needed)
claude mcp add --transport stdio enrichr -- npx -y enrichr-mcp-server

Example:

> Run enrichment analysis on this gene list: BRCA1, TP53, ATM, CHEK2, PALB2
> Show me the top GO Biological Process and KEGG pathway terms.

Holy Bio MCP (Longevity/Aging Research + gget)

Aggregates multiple bio sub-servers including gget (popular Python genomics library), OpenGenes (aging/longevity), and BioThings APIs.

# See: https://lobehub.com/mcp/longevity-genie-holy-bio-mcp
# NOTE: "holy-bio-mcp" is NOT available on PyPI as of Feb 2026.
# Check the link above for current install instructions. You may need to
# install from source:
git clone https://github.com/longevity-genie/holy-bio-mcp.git
cd holy-bio-mcp && pip install -e .
claude mcp add --transport stdio holybio -- holy-bio-mcp

4. MCP Server Reference Table

Complete reference of all biological MCP servers discussed in this guide.

Official Anthropic Life Sciences Connectors

Server	Transport	What It Provides	Install	Auth Required
PubMed	HTTP	Millions of biomedical articles (Anthropic-hosted)	One command	No
bioRxiv / medRxiv	HTTP	260k+ preprints (biology + medicine)	One command	No
ChEMBL	HTTP	2.4M+ bioactive compounds, targets, activities	One command	No
ClinicalTrials.gov	HTTP	500k+ clinical studies (NIH/NLM)	One command	No
Synapse.org	HTTP	Biomedical datasets, project structure	One command	Yes (OAuth2)
Scholar Gateway (Wiley)	HTTP	Scientific research papers	One command	Yes (CONNECT SSO)
Owkin Pathology Explorer	HTTP	H&E pathology slide analysis (TCGA)	One command	Yes
Medidata	HTTP	Clinical trial platform (Rave EDC, site ranking)	One command	Yes (Medidata acct)
10x Genomics Cloud	stdio (MCPB)	Single cell + spatial analysis	Extension/plugin	Yes (access token)
ToolUniverse	stdio	600+ scientific tools (ML, APIs, datasets)	pip install	Optional API keys

All by deepsense.ai (bioRxiv, ChEMBL, ClinicalTrials.gov), Sage Bionetworks (Synapse), Wiley (Scholar Gateway), Owkin, Medidata, 10x Genomics, and MIMS Harvard (ToolUniverse).

Community & Third-Party Servers

Server	Transport	Databases/Tools	Install Complexity	Best For
BioContextAI	HTTP	20+ databases (UniProt, AlphaFold, Ensembl, KEGG, GO, etc.)	One command	General bio research (start here)
BioRender	HTTP	50k+ scientific icons/templates	One command (needs account)	Figures, posters, graphical abstracts
Benchling	HTTP	Your lab's ELN, registry, inventory	Admin setup required	Labs using Benchling
BioMCP	stdio	PubMed, ClinicalTrials.gov, genomic variants	pip install	Literature + clinical + variants
UniProt MCP	stdio	UniProt (26 tools)	git clone + npm + build	Deep protein research
AlphaFold MCP	stdio	AlphaFold DB	git clone + npm + build	Predicted structures
PDB MCP	stdio	Protein Data Bank	git clone + npm + build	Experimental structures
Reactome MCP	stdio	Reactome pathways	git clone + npm + build	Pathway analysis
Enrichr MCP	stdio	GO, KEGG, Reactome, WikiPathways, MSigDB	npx (one command)	Gene set enrichment
PubMed MCP	stdio	PubMed/NCBI	npx (one command)	Literature search
Paper Search	stdio	arXiv + PubMed + bioRxiv + medRxiv	pip install	Multi-source lit search
bio-mcp-blast	stdio	NCBI BLAST	git clone + pip + conda	Sequence similarity
bio-mcp-samtools	stdio	SAMtools	git clone + pip + conda	BAM/SAM processing
bio-mcp-bwa	stdio	BWA	git clone + pip + conda	Read alignment
bio-mcp-seqkit	stdio	SeqKit	git clone + pip + conda	Sequence utilities
bio-mcp-fastqc	stdio	FastQC/MultiQC	git clone + pip + conda	Sequencing QC
Holy Bio MCP	stdio	gget, OpenGenes, BioThings	git clone + pip (not on PyPI)	Genomics + aging research
Protocols.io	stdio	Protocols.io	pip install	Protocol management

5. Future-Proofing: Stay Up to Date

The MCP ecosystem for biology is growing fast. Here's how to keep your setup current.

Automated Discovery: Check for New Servers

Create a skill that checks for new MCP servers. Save as .claude/skills/check-bio-mcp/SKILL.md:

---
name: check-bio-mcp
description: Check for new biological MCP servers and updates
---

Search the web for new MCP servers relevant to biological research:

1. Check these registries for new entries:
   - https://biocontext.ai/registry
   - https://mcpmed.org
   - https://github.com/punkpeye/awesome-mcp-servers
   - https://github.com/anthropics/life-sciences
   - https://glama.ai/mcp/servers (filter: biology/science)

2. Search GitHub for recent repos:
   - "mcp server" biology created:>2026-01-01
   - "mcp server" bioinformatics created:>2026-01-01
   - "mcp server" genomics created:>2026-01-01

3. Compare findings against our currently installed servers (run `claude mcp list`)

4. For any NEW servers found:
   - Summarize what they do
   - Note install complexity
   - Rate relevance to functional genomics / deep mutational scanning
   - Provide the install command

5. Check for updates to servers we already use:
   - BioContextAI: check GitHub releases
   - BioMCP: check PyPI for new version
   - bio-mcp suite: check for new tool wrappers

Usage: /check-bio-mcp — run this monthly to stay current.

Key Registries to Watch

Registry	URL	What It Tracks
BioContextAI Registry	biocontext.ai/registry	Curated bio database MCP servers
MCPMed Hub	mcpmed.org	Expert-validated bioinformatics MCPs
Awesome MCP Servers	GitHub: punkpeye/awesome-mcp-servers	Community-curated list of all MCP servers
Anthropic Life Sciences	GitHub: anthropics/life-sciences	Official Anthropic plugins for science
Glama MCP Directory	glama.ai/mcp/servers	Searchable MCP server directory
bio-mcp GitHub Org	github.com/bio-mcp	Bioinformatics CLI tool wrappers

Emerging Projects to Watch

Science Context Protocol (SCP)

An extension of MCP specifically designed for multi-lab scientific workflows. Adds:

Richer experiment metadata
Centralized hub (vs. MCP's peer-to-peer model)
Workflow orchestration through an experiment flow API
Lab device integration with standardized drivers

Their Internal Discovery Platform currently provides 1,600+ interoperable tools, with biology making up 45.9% of them. Not yet publicly available but worth tracking.

BioinfoMCP (Automated Tool Conversion)

An academic project (arXiv: 2510.02139) that automatically converts bioinformatics CLI tools into MCP servers by analyzing their documentation and CLI interfaces. They've converted 38 tools with 94.7% success rate. If this matures, it could make every bioinformatics tool instantly MCP-compatible.

CZI / CZ Biohub

No public MCP integration as of February 2026. CZI tools like CZ ID (metagenomic pathogen detection) and cellxgene (single-cell data explorer) are good candidates for future MCP wrappers. Worth checking periodically.

GitHub Watch Commands

Set up notifications for key repositories:

# Watch for new releases from key MCP projects
gh repo set-default biocontext-ai/knowledgebase-mcp
gh api repos/biocontext-ai/knowledgebase-mcp/subscription -f subscribed=true

# Star and watch the bio-mcp organization repos
gh api repos/bio-mcp/bio-mcp-blast/subscription -f subscribed=true

# Watch the awesome-mcp-servers list for new additions
gh api repos/punkpeye/awesome-mcp-servers/subscription -f subscribed=true

6. Recommended Project Structure

Here's the complete .claude/ directory structure for a bio research project:

your-project/
├── .claude/
│   ├── settings.json          # Hooks, permissions, MCP config
│   ├── skills/
│   │   ├── fix-bug/
│   │   │   └── SKILL.md       # /fix-bug [function]
│   │   ├── run-tests/
│   │   │   └── SKILL.md       # /run-tests
│   │   ├── search-lit/
│   │   │   └── SKILL.md       # /search-lit [topic]
│   │   ├── lookup-protein/
│   │   │   └── SKILL.md       # /lookup-protein [uniprot-id]
│   │   ├── enrichment/
│   │   │   └── SKILL.md       # /enrichment [gene-list]
│   │   └── check-bio-mcp/
│   │       └── SKILL.md       # /check-bio-mcp (future-proofing)
│   └── agents/
│       ├── test-runner.md      # Run tests, diagnose failures
│       └── lit-search.md       # Search bio literature
├── CLAUDE.md                   # Project context and rules
├── calculator.py
├── test_calculator.py
└── ...

Example settings.json (Full BioClaude Config)

{
  "permissions": {
    "allow": [
      "Bash(pytest:*)",
      "Bash(python:*)",
      "Bash(git:*)",
      "Bash(conda:*)",
      "Bash(pip:*)"
    ]
  },
  "hooks": {
    "PostToolUse": [
      {
        "matcher": "Edit|Write",
        "hooks": [
          {
            "type": "command",
            "command": "cd $CLAUDE_PROJECT_DIR && python -m pytest -q 2>&1 | tail -5"
          }
        ]
      }
    ],
    "SessionStart": [
      {
        "hooks": [
          {
            "type": "command",
            "command": "echo 'BioClaude active. MCP servers:' && claude mcp list 2>/dev/null || echo '(run claude mcp list to check)'"
          }
        ]
      }
    ]
  },
  "mcpServers": {
    "biocontext": {
      "transport": "http",
      "url": "https://mcp.biocontext.ai/mcp/"
    },
    "biorender": {
      "transport": "http",
      "url": "https://mcp.services.biorender.com/mcp"
    },
    "biomcp": {
      "transport": "stdio",
      "command": "biomcp",
      "args": ["run"]
    }
  }
}

7. Troubleshooting

MCP server won't connect

# Check if the server is configured
claude mcp list

# Test a remote server URL directly
curl -s https://mcp.biocontext.ai/mcp/ | head -20

# For stdio servers, test the command runs
biomcp run --help

# Remove and re-add a server
claude mcp remove biocontext
claude mcp add --transport http biocontext https://mcp.biocontext.ai/mcp/

"Tool not found" errors

MCP tools only appear after the server successfully connects. If tools aren't showing:

Restart Claude Code (/exit then claude)
Check server logs: claude mcp logs <server-name>
Verify the server process is running (for stdio servers)

Hooks not firing

Verify .claude/settings.json is valid JSON (use python -m json.tool)
Check the matcher regex matches the tool name exactly
Test the hook command manually in your terminal
Check that the hook script is executable

Permission errors with bio tools

# Ensure conda environment is activated before starting Claude
conda activate biotools
claude

# Or add conda activation to your session start hook

Rate limiting on remote MCP servers

BioContextAI's hosted server is subject to fair use limits. For heavy use:

# Switch to local installation
pip install biocontext-kb
claude mcp remove biocontext
claude mcp add --transport stdio biocontext-local -- uvx biocontext_kb@latest

FilesExpand file tree

Advanced Tools Setup.MD

Latest commit

History

Advanced Tools Setup.MD

File metadata and controls

Advanced Claude Code Tools Setup

Table of Contents

1. Advanced Claude Code Features

Hooks

Skills (Custom Slash Commands)

Custom Agents (Subagents)

Extended Thinking

Permissions Whitelisting

2. MCP (Model Context Protocol) Overview

3. BioClaude: Full Bio Research Setup

Quick Start (5 minutes)

Official Anthropic Life Sciences Connectors

Scientific Research & Discovery

Clinical Operations & Regulatory

Local Extensions (Desktop/stdio)

Tier 1: Core Bio Databases

BioContextAI (Recommended Starting Point)

Individual Database Servers (Augmented Nature Suite)

Tier 2: Lab Platforms

BioRender

Benchling

Protocols.io

Tier 3: Bioinformatics CLI Tools

Tier 4: Literature & Enrichment

PubMed MCP Server

BioMCP (GenomOncology)

Enrichr MCP Server

Holy Bio MCP (Longevity/Aging Research + gget)

4. MCP Server Reference Table

Official Anthropic Life Sciences Connectors

Community & Third-Party Servers

5. Future-Proofing: Stay Up to Date

Automated Discovery: Check for New Servers

Key Registries to Watch

Emerging Projects to Watch

Science Context Protocol (SCP)

BioinfoMCP (Automated Tool Conversion)

CZI / CZ Biohub

GitHub Watch Commands

6. Recommended Project Structure

Example settings.json (Full BioClaude Config)

7. Troubleshooting

MCP server won't connect

"Tool not found" errors

Hooks not firing

Permission errors with bio tools

Rate limiting on remote MCP servers

Further Reading