faststack-mcp
Local-first, read-only MCP server for Claude Code that indexes and searches full-stack projects (FastAPI + React + PostgreSQL) with SQLite-backed FTS, cross-reference graph, and file-watcher auto-reload.
README
faststack-mcp
Local-first, read-only MCP server for Claude Code. Indexes and searches full-stack projects (FastAPI + React + PostgreSQL) with SQLite-backed FTS, cross-reference graph, and file-watcher auto-reload — so Claude reads only what it needs instead of entire files.
Quick Start
Step 1 — Install
# From local source (development)
cd C:\path\to\faststack-mcp
pip install -e .
# From GitHub
pip install "git+https://github.com/Evaan-devlops/faststack-mcp.git"
# With optional extras
pip install -e ".[watch]" # file watcher (watchfiles)
pip install -e ".[postgres]" # PostgreSQL backend (psycopg2-binary)
pip install -e ".[all]" # both
Already have watchfiles / psycopg2 globally? Nothing extra to install — they're picked up automatically.
Step 2 — Add to Claude Code
{
"mcpServers": {
"faststack": {
"command": ".venv\\Scripts\\python.exe",
"args": ["-m", "faststack_mcp"]
}
}
}
Step 3 — Index your project
index_folder("/path/to/your/project") → project_id + counts
index_folder("/path/to/project", watch=True) → + auto-reindex on file changes
Step 4 — Explore
get_project_outline(project_id) → structure at a glance (~17 tokens)
search_symbols(project_id, "createUser") → find functions, routes, hooks, models
get_symbol(project_id, symbol_id) → exact code snippet only
find_references(project_id, "get_db") → all files that import / call a symbol
get_file_context(project_id, "db.py", 120) → 80 lines around line 120
get_file_outline(project_id, "main.py") → all symbols in one file
search_text(project_id, "HTTPException") → fallback full-text search
get_file_tree(project_id) → directory tree
list_projects() → all indexed projects
invalidate_cache(project_id) → force full re-index
get_token_usage() → cost + per-session averages
Tools reference (12 tools)
| Tool | Purpose | Key params |
|---|---|---|
index_folder |
Scan & cache a project | path, force, watch, include_env_files |
list_projects |
Browse cached projects | — |
search_symbols |
Primary nav — ranked FTS symbol search | query, kind, language, include_synthetic |
get_symbol |
Read exact code snippet by symbol_id | symbol_id |
find_references |
New — import graph + text scan for a symbol | symbol_name, limit |
get_file_context |
New — read N lines around a line number | file_path, around_line, radius=40 |
search_text |
Full-text search across all files | query, limit |
get_project_outline |
Grouped structure overview | sections |
get_file_tree |
Directory tree of indexed files | — |
get_file_outline |
All symbols in one file | file_path |
invalidate_cache |
Remove cached index | project_id |
get_token_usage |
Session cost + tool breakdown | last_n, session_id |
find_references — dependency impact analysis
find_references(project_id, "get_db")
→ {
symbol_name: "get_db",
files_found: 3,
total_usages: 3,
references: [
{ file_path: "users.py", usages: [{ line: 5, context: "from db import get_db" }] },
{ file_path: "auth.py", usages: [{ line: 12, context: "db = get_db()" }] },
{ file_path: "orders.py", usages: [{ line: 8, context: "from db import get_db" }] }
]
}
Use before modifying a shared symbol to understand blast radius. Uses the import graph built
during index_folder (fast path) with a whole-word regex scan as fallback.
get_file_context — surrounding code window
get_file_context(project_id, "src/db.py", around_line=120, radius=40)
→ lines 80–160, line 120 marked with >>>
Use when get_symbol is too narrow — e.g. understanding the code around a symbol boundary,
inspecting a migration function, or reading error-handling context.
Storage backends
Controlled by FASTSTACK_STORAGE environment variable.
| Value | Default | When to use |
|---|---|---|
sqlite |
Yes | Local dev, any project size. Single ~/.faststack-mcp/faststack.db with FTS5. |
postgres |
No | Shared team index, large projects, or when you want the index in your project DB. Requires DATABASE_URL. |
json |
No | Backward compat with pre-v0.2 caches. No FTS or reference graph. |
SQLite (default)
No configuration needed. Features:
- FTS5 virtual table on symbol name + qualified_name + signature
- Combined FTS5 + LIKE fallback (handles
get_db,create_userunderscore names) refstable forfind_referenceslookups- WAL mode — concurrent reads while indexing
PostgreSQL
export FASTSTACK_STORAGE=postgres
export DATABASE_URL=postgresql://user:pass@localhost:5433/mydb
Features:
GENERATED ALWAYS AStsvector column + GIN index (zero-maintenance FTS)ts_rankrelevance ordering + ILIKE supplement- Tables prefixed
faststack_*(no conflicts with your app tables) - Batch inserts via
execute_values(500 rows/page) faststack_referencestable forfind_references
psycopg2-binarymust be installed (pip install psycopg2-binaryorpip install faststack-mcp[postgres]).
File watcher (auto-reindex)
# Start watcher alongside indexing
index_folder("/path/to/project", watch=True)
- Uses
watchfiles(install:pip install watchfilesorpip install faststack-mcp[watch]) - 1.5 s debounce — rapid saves are batched before re-index fires
- Background daemon thread — does not block Claude
- Falls back silently if
watchfilesis not installed
Cross-reference graph
Built automatically during index_folder. Tracks every import and from X import Y
statement across Python and TypeScript/JavaScript files.
The graph is stored in the index and queried by find_references. It enables:
- "What calls / imports this symbol?" in one tool call
- Impact analysis before refactoring shared utilities
- Understanding FastAPI dependency injection (
Depends(get_db))
Search — include_synthetic behaviour
Synthetic symbols (auto-generated model fields, config keys) are shown or hidden depending on context:
| Query kind | Default |
|---|---|
pydantic_model, model, type, interface |
Shown (fields are useful) |
config_key, config_file, tsconfig, json_key |
Shown |
| Everything else | Hidden |
Override explicitly: search_symbols(project_id, "User", include_synthetic=True)
Token usage analytics
get_token_usage(last_n=20)
→ {
totals: { input_tokens, output_tokens, estimated_cost_usd },
averages: { input_tokens_per_session, cost_usd_per_session },
tool_breakdown: { get_symbol: { calls: 18, est_saved: 136800 }, ... },
estimated_tokens_saved_by_mcp: 209000
}
Requires the Stop hook in ~/.claude/settings.json to log session data.
Supported file types
Source: .py .ts .tsx .js .jsx .sql
Config / meta: .json .jsonl .toml .yaml .yml .ini
Named files: package.json pyproject.toml tsconfig.json tsconfig.app.json
tsconfig.node.json vite.config.* alembic.ini eslint.config.js
tailwind.config.js tailwind.config.ts manifest.json chunks.jsonl
index.faiss bunfig.toml .env.example
Skipped dirs: .git node_modules .next dist build coverage .venv venv
__pycache__ .mypy_cache .pytest_cache .idea .vscode .claude
Sensitive files skipped: .env .env.* *.pem *.key *.p12 id_rsa id_ed25519
(.env* files opt-in via include_env_files=True)
Parsers
| Language | Parser | Extracts |
|---|---|---|
| Python / FastAPI | ast |
functions, classes, routes, pydantic models, services, repos, decorators, Depends() |
| TypeScript / React | tree-sitter + regex fallback |
components, hooks, types, interfaces, backend routes/services/repos |
| SQL | regex | tables, views, indexes, functions, procedures, triggers |
| JSON / JSONL | json.loads |
root keys, RAG chunk fields, manifest metadata |
| Config | tomllib / json / yaml / ini |
package scripts/deps, tsconfig, vite, eslint, alembic |
| Env | regex (masked) | variable names — values stored as ***MASKED*** |
| Tailwind | heuristics | config keys, className usage |
Security
- All file reads path-confined to the indexed project root
- Symlink escape checks on every lookup
- Binary file detection (null-byte scan)
- File size cap (default 2 MB)
- Sensitive file patterns skipped by default
- No write or shell operations on project files
- SQLite cache isolated at
~/.faststack-mcp/
Installation options
# Local dev
pip install -e "C:\path\to\faststack-mcp"
# GitHub
pip install "git+https://github.com/Evaan-devlops/faststack-mcp.git"
# With extras
pip install -e ".[watch]" # watchfiles — file watcher
pip install -e ".[postgres]" # psycopg2-binary — PostgreSQL backend
pip install -e ".[all]" # both
Environment variables
| Variable | Default | Description |
|---|---|---|
FASTSTACK_STORAGE |
sqlite |
Storage backend: sqlite, postgres, json |
DATABASE_URL |
— | PostgreSQL DSN (required when FASTSTACK_STORAGE=postgres) |
Development
python -m venv .venv
.venv\Scripts\python -m pip install -e .
python -m py_compile src/faststack_mcp/server.py # syntax check
pytest tests/
Env parsing
.env.exampleindexed by default with masked values (***MASKED***).envand.env.*skipped unlessinclude_env_files=True- Raw values never persisted
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.