75 MCP Tools
for AI Coding Agents
A toolkit for code intelligence, context memory, and LSP navigation — all served by the unified MCP server over stdio (devscriptor mcp). Works with Claude Desktop, Cursor, Cline, and any MCP-compatible agent.
Core tools operate on the local SQLite knowledge graph built by devscriptor scan. They require no language server.
Scanning & Analysis Jobs
scan_codebase
Scan a codebase and build the entity/relationship database
{
"path": "/path/to/project",
"extensions": ["rs", ".toml"],
"exclude_patterns": ["**/tests/**", "*_test.rs"]
}analyze_code
Run code analysis on a path (dead code, complexity, smells)
start_analysis
Start a background analysis job
get_analysis_progress
Get the progress of a running analysis job
list_analysis_jobs
List all analysis jobs
cancel_analysis
Cancel a running analysis job
cleanup_old_jobs
Clean up old completed jobs
Statistics & Health
get_statistics
Database statistics: total files, entities, and relationships (scopable by path)
health_check
Check server health (database, jobs, version)
Search & Embeddings
search_entities
Search for entities in the database
search_functions
Search for functions by name
search_classes
Search for classes/types by name
fuzzy_search
Fuzzy search across all entities
semantic_search
Semantic (vector) search using embeddings. Embeds matching candidates on demand, so it works right after a scan with no separate embed step — run generate_embeddings first to pre-build the whole index for larger codebases
{
"query": "function to read and parse config files",
"entity_type": "function",
"limit": 10
}generate_embeddings
Generate and store embedding vectors for entities missing them (idempotent); pre-builds the index semantic_search uses
{
"force": false,
"batch_size": 16
}Code Quality Analysis
find_dead_code
Find potentially unused code
analyze_complexity
Analyze code complexity
Entities & Relationships
get_entity_details
Get detailed information about an entity
{
"entity_type": "function",
"name": "parse_file",
"file_path": "src/parser.rs"
}get_relationships
Get relationships for an entity (calls, inherits, implements, …)
Context Graph
add_context_note
Add a note, optionally linked to code entities and concepts
search_context
Search context notes (keyword / semantic / hybrid + filters)
get_related_context
Get notes linked to an entity type
prune_context
Prune (archive) old context notes
get_context_statistics
Get context graph statistics
File Context & Edit Validation
describe_file
Get a description of a file's contents
describe_modification
Describe what a modification would do
get_file_context
Get context for a file
validate_file_edit
Validate a file edit with context
query_with_context
Query with context awareness
Real-time Watcher
start_watching
Start watching a directory and auto-update the graph
{
"path": "/path/to/project",
"debounce_ms": 400
}stop_watching
Stop the real-time watcher
get_watch_status
Report watch status, re-scan count, and last re-scan stats
Connect Your MCP Client
Point your client at the devscriptor binary on your PATH with args: ["mcp"]. It serves all 75 tools over JSON-RPC on stdio — there is no network port.
Claude Desktop / Claude Code / Cline
{
"mcpServers": {
"devscriptor": {
"command": "devscriptor",
"args": ["mcp"]
}
}
}For debug logs on stderr use "args": ["--verbose", "mcp"]. Setting RUST_LOG has no effect — it is not read.
Cursor
{
"mcpServers": {
"devscriptor": {
"command": "devscriptor",
"args": ["mcp"]
}
}
}Identical on every platform — Devscriptor ships one executable answering to one name (devscriptor.exe on Windows).
Works with Any MCP Client
All 75 tools are available through the Model Context Protocol — compatible with Claude Desktop, Cursor, Cline, and any other MCP-compatible agent.