Softagram Analyzer MCP

Run Softagram code analysis in a Docker container and query architectural intelligence via Model Context Protocol (MCP).

Claude (or any MCP-compatible AI assistant) can analyze your repositories and answer questions like:

• "What depends on this class?"
• "What would break if I change this function?"
• "Show me the structure of this module"

Quick Start

Automated setup (recommended)

The setup script walks you through login, repository selection, MCP configuration, and verification — all in one command:

./setup.sh

Manual setup

Click to expand manual steps

0. Log in to the Softagram registry

You need credentials provided by Softagram. Contact support@softagram.com if you don't have them.

docker login registry.softagram.com

1. Start the analyzer

Edit docker-compose.yml to mount your repositories (see Mounting Your Code), then:

docker compose up -d

Or without Compose:

docker run -d --name softagram-analyzer \
  --tmpfs /tmp:size=2G \
  -v ~/code/my-project:/input/my-project:ro \
  -p 8008:8008 \
  registry.softagram.com/softagram-analyzer-mcp:latest

2. Configure Claude Code

Add Softagram to your MCP configuration. If you don't have ~/.claude/mcp.json yet:

cp mcp.json.example ~/.claude/mcp.json

If you already have ~/.claude/mcp.json with other servers, merge the "softagram" entry into your existing mcpServers object — don't overwrite the file.

For project-specific configuration, copy to .mcp.json in the project root instead.

3. Verify it works

# Check the container is running
docker ps | grep softagram-analyzer

# Check the MCP server responds
curl -s -o /dev/null -w "%{http_code}" http://localhost:8008/mcp
# Any response (200, 405, etc.) means the server is up.
# "000" or connection refused means it's still starting — wait a few seconds.

4. Use with Claude

Start a new Claude Code session (needed to pick up MCP config), then:

You: "Analyze my project at /input/my-project"
You: "What depends on the UserService class?"
You: "What would break if I change the validate function?"
You: "Show me the architecture of the auth module"

Mounting Your Code

The container needs access to the repositories you want to analyze. Mount them under /input:

# docker-compose.yml — add your repos under volumes:
services:
  analyzer:
    volumes:
      - ~/code/my-frontend:/input/my-frontend:ro
      - ~/code/my-backend:/input/my-backend:ro

Or with docker run:

docker run -d --name softagram-analyzer \
  -v ~/code/my-frontend:/input/my-frontend:ro \
  -v ~/code/my-backend:/input/my-backend:ro \
  -p 8008:8008 \
  registry.softagram.com/softagram-analyzer-mcp:latest

Repositories are mounted read-only (:ro) — the analyzer never modifies your code.

You can also analyze public repositories by URL without mounting:

"Analyze https://github.com/expressjs/express"

Persisting Results

Analysis results are stored inside the container at /tmp/analysis-outputs. To keep them across container restarts:

# docker-compose.yml
services:
  analyzer:
    volumes:
      - ./analysis-outputs:/tmp/analysis-outputs

Available Tools

Once connected, Claude has access to these tools:

Tool	What it does
analyze_repo	Run full code analysis on a repository
get_analysis_status	List completed analyses
load_model	Load an analysis model for querying
search_elements	Find classes, functions, files by name
get_dependencies	What does X depend on? What depends on X?
get_structure	Explore the code hierarchy
analyze_change_impact	What breaks if I change X?
download_model	Export the analysis model

Example Workflow

1. "Analyze /input/my-backend"
   → runs full analysis, produces architectural model

2. "What does the OrderService depend on?"
   → loads model, searches for OrderService, queries outgoing dependencies

3. "What would break if I change the PaymentGateway interface?"
   → traces all direct and transitive dependents, grouped by file

4. "Show me the structure of the api/ directory"
   → displays file/class/function hierarchy

Configuration

Environment Variables

Variable	Default	Description
MCP_PORT	8008	Port the MCP server listens on
MCP_TRANSPORT	streamable-http	Protocol: streamable-http or sse

MCP Transport

The default transport is Streamable HTTP — the current MCP standard for remote servers. It uses stateless HTTP POST requests and works well with proxies and load balancers.

Endpoint: http://localhost:8008/mcp

If your MCP client only supports SSE (legacy), set the environment variable:

# docker-compose.yml
services:
  analyzer:
    environment:
      - MCP_TRANSPORT=sse

SSE endpoint: http://localhost:8008/sse

Changing the Port

# docker-compose.yml
services:
  analyzer:
    ports:
      - "9090:8008"

Update mcp.json accordingly:

{
  "mcpServers": {
    "softagram": {
      "type": "streamable-http",
      "url": "http://localhost:9090/mcp"
    }
  }
}

System Requirements

• Docker (Docker Desktop or Docker Engine)
• ~4 GB disk space for the image
• ~2 GB RAM recommended
• Network access for cloning Git repositories (optional)

Supported Languages

Softagram analyzes dependencies, structure, and metrics for:

Python, Java, Kotlin, JavaScript, TypeScript, C#, F#, C/C++, Go, Dart, Ruby, PHP, Scala, Groovy, Swift, Objective-C, Haskell, Clojure, COBOL, MATLAB, and more.

Verify It Works

After starting the container, check that the MCP server is responding:

curl -s -o /dev/null -w "%{http_code}" http://localhost:8008/mcp

• Any HTTP status (200, 405, etc.) — server is running
• 000 / connection refused — server is still starting, wait a few seconds and retry
• No response after 60 seconds — check docker logs softagram-analyzer

Then start a new Claude Code session and ask Claude to list available tools — you should see the Softagram tools (analyze_repo, get_dependencies, etc.).

Troubleshooting

Container won't start:

docker logs softagram-analyzer

"Connection refused" on port 8008: Wait a few seconds after starting — the server needs time to initialize. Verify with: curl -s -o /dev/null -w "%{http_code}" http://localhost:8008/mcp

Analysis takes a long time: First analysis of a large codebase can take several minutes. Subsequent analyses are faster due to caching.

Image not found:

docker pull registry.softagram.com/softagram-analyzer-mcp:latest

MCP tools not visible in Claude Code: Make sure ~/.claude/mcp.json (or .mcp.json) contains the "softagram" server entry, and that you started a new session after adding it.

License

This software is proprietary. See LICENSE for details. Usage requires a valid Softagram license.

Support

• Documentation: docs.softagram.com
• Email: support@softagram.com