start_conversationLABSCO SUMMARY
Routing between models, not replacing one with another.
The premise is that these are heterogeneous services rather than competitors: one is good at tight terminal loops and multi-file edits, the other at a single enormous sweep over logs, traces and code at once. The tools are shaped around that handoff — you tell it what you already tried and where you got stuck, which is the context a fresh model normally lacks.
What it is
An escalation route between two models. Claude Code keeps the local, incremental work; when a problem outgrows that — gigabytes of traces, failures spanning ten services, twenty hypotheses to test — this server packages the context and hands it to Gemini 2.5 Pro, then returns the result for implementation.
What you get
escalate_analysis— the main handoff: what was attempted, what was found, where it stuck, and the files in scope, with an analysis type ofexecution_trace,cross_system,performanceorhypothesis_test, a depth level from 1 to 5 and an optional time budgetstart_conversation,continue_conversation,finalize_conversation,get_conversation_status— a multi-turn dialogue between the two models, closing with a summary formatteddetailed,conciseoractionabletrace_execution_path— follow execution from a file and line with data flow included, to a configurable depthcross_system_impact— model breaking, performance and behavioural effects of a change across service boundariesperformance_bottleneck— N+1 query patterns, memory leaks and algorithmic hot spots, at a profile depth you choosehypothesis_test— state a theory about the code and a way to test it, and get evidence-based validation back
Requirements
Node.js 18 or later and a Gemini API key from Google AI Studio in GEMINI_API_KEY. Cloned and built with npm install and npm run build, then launched as node dist/index.js. Two practical rules the troubleshooting section spells out: file paths passed to tools must be absolute, and every parameter name is snake_case (claude_context, not claudeContext). Your code is sent to Google's Gemini API, so read their data policies before pointing it at a private repository.
Setup effort
Build from source — git clone https://github.com/evalops/deep-code-reasoning-mcp.git
USE CASES
01
Correlating a failure across many services
Escalate with the trace files and service names in scope and an execution_trace or cross_system analysis type, and let the larger context window hold the whole timeline.
02
Testing a theory you cannot cheaply reproduce
State the hypothesis and the test approach, and get back validation with the evidence rather than an opinion.
03
Chasing a regression that profiling did not explain
Point it at the hot path and ask for bottleneck analysis, then keep the conversation going until the answer is actionable.
9 tools
Listed in project docs. Your client may show a different list.
continue_conversationfinalize_conversationget_conversation_statusescalate_analysistrace_execution_pathcross_system_impactperformance_bottleneckhypothesis_testBEFORE YOU INSTALL
What to know first
Tools you get
9 tools named in the project's documentation.
Requires
- Node.js 18
- GEMINI_API_KEY in your environment
Where it runs
Both documented clients run the server on your machine.
INSTALLATION
Connect your client
You clone and build this one yourself.
Server setup
TERMINAL
git clone https://github.com/evalops/deep-code-reasoning-mcp.git
cd deep-code-reasoning-mcp
npm install
npm run buildMCP Clients
Claude Desktop
CLAUDE DESKTOP · JSON
{
"mcpServers": {
"deep-code-reasoning": {
"command": "node",
"args": [
"<YOUR_PATH_TO_DEEP_CODE_REASONING_MCP>/dist/index.js"
],
"env": {
"GEMINI_API_KEY": "<YOUR_GEMINI_API_KEY>"
}
}
}
}Cursor
CURSOR · JSON
{
"mcpServers": {
"deep-code-reasoning-mcp": {
"command": "node",
"args": [
"<YOUR_PATH_TO_DEEP_CODE_REASONING_MCP>/dist/index.js"
],
"env": {
"GEMINI_API_KEY": "<YOUR_GEMINI_API_KEY>"
}
}
}
}
