Integrations

Vibe Analyzer provides an MCP server that AI assistants can connect to via the Model Context Protocol. Once connected, the model gains 11 tools for searching code and documentation.

How It Works from the User’s Perspective

The user communicates with the AI assistant in natural language. The model decides which tool to call. Examples from real testing scenarios:

Code Search

User Query	Tool	What Happens
“Find `add` functions in the samples project”	`search_by_code_functions`	Searches for functions with `add` in the signature, returns files and signatures
“What classes are in samples?”	`search_by_code_classes`	Returns all classes, structs, interfaces
“Show all enums in samples”	`search_by_code_variables`	Enums are also searched through this tool
“What libraries are used in samples?”	`search_by_code_imports`	List of all imports in the project
“List files that have the `MAX_VALUE` constant”	`search_by_code_variables`	Search by constant name

File Viewing

User Query	Tool
“Show the contents of `src/main.rs`”	`get_file_content`
“Show the structure of `main.py`”	`get_file_ast`
“What functions are in `src/main.rs`?”	`get_file_ast`
“Open `utils.py`”	`get_file_content`

Documentation and Knowledge Base Search

User Query	Tool
“Who is Zizikosh?”	`search_documentation`
“Tell me about Kukyrbur’s abilities”	`search_documentation`
“Find Python coding guidelines”	`search_documentation`
“Show the release process”	`search_documentation`
“Find the code review checklist”	`search_documentation`

User Query	Tool
“What projects are in the database?”	`show_projects`
“Show the tree of the samples project”	`show_tree`
“How many files are in knowledge?”	`show_stats`
“Show overall statistics for all projects”	`show_stats`

Administration

User Query	Tool
“Update the index”	`admin_sync`
“Reindex projects”	`admin_sync`

How to Phrase Queries

The model understands queries in natural language. You don’t need to use exact tool names — plain language is enough.

Good:

“Find add functions in the samples project”
“What classes are in samples?”
“Show the contents of src/main.rs”
“Who is Zizikosh?”

Unnecessary (the model will understand via AliasHandler anyway, but it’s better to avoid):

“Call search_by_code_functions with query=add”
“Use the get_file_content tool for path=src/main.rs”

Important Notes

Project names — you can use the full path or directory name: "samples" or "/path/to/samples"
File paths — relative to the project root: "src/main.rs", partial matching is supported
Result limit — default 3, maximum 10. If the model requests “all”, the limit is automatically raised
One call is enough — the model is trained to respond after a single tool call, no need to ask again

Connecting to Open WebUI

Start the MCP server:
```
vibe-analyzer serve start
```
In Open WebUI settings, add a new MCP server:
- URL: http://localhost:9020
- Transport: Streamable HTTP
Tools appear automatically

Connecting to Claude Desktop

Add to the configuration:

{
  "mcpServers": {
    "vibe-analyzer": {
      "url": "http://localhost:9020",
      "transport": "streamable-http"
    }
  }
}

MCP Protocol

Supported versions: 2024-11-05, 2025-03-26, 2025-06-18, latest. Configured in the settings:

{
  "mcp": {
    "host": "127.0.0.1",
    "port": 9020,
    "protocol": "latest"
  }
}

Security

Server without authentication — for trusted networks or localhost
Default host 127.0.0.1 (local only)
0.0.0.0 — for access from Docker containers or other machines
Server only reads data, admin_sync is the only tool that triggers background indexing

Keyboard shortcuts

Vibe Analyzer