Built in Tools

Built-in Tools

Synapse agents are powerful because of what they can do. Every tool runs as a separate MCP process — isolated, composable, and safe.

---

Native Tool Servers

These start automatically when Synapse starts. No configuration required.

Tool	What It Does
Sandbox	Execute Python code in an isolated Docker container (512 MB RAM, 1 CPU). Pre-loaded with pandas, numpy, matplotlib, scikit-learn, requests, and more. Can read/write files in the persistent vault.
Vault	Persistent file storage for agents. Create, read, update, patch, and list files across sessions. JSON deep-merge, text find-replace, and directory listing built in.
SQL Agent	Connect to any database (PostgreSQL, MySQL, SQLite). List tables, introspect schemas, run read queries. Supports any SQLAlchemy-compatible connection string.
Browser	Full browser automation via Playwright MCP. Navigate pages, click, fill forms, take screenshots, extract content. Powered by Chromium.
PDF Parser	Extract text and tables from any PDF by URL. Tables converted to Markdown. Page-by-page extraction.
Excel Parser	Parse .xlsx files from URL. Multi-sheet support. Converts all sheets to Markdown tables.
Web Scraper	Powerful web scraping powered by crawl4ai. Scrape any URL to clean markdown, extract structured data with CSS schemas, crawl multiple URLs in parallel, capture screenshots, handle infinite-scroll pages, and run multi-step authenticated sessions. Built-in stealth mode bypasses anti-bot protections — works on LinkedIn, financial sites, and JavaScript-heavy pages.
Collect Data	Generate dynamic forms that pause execution and collect user input. Supports text, number, email, date, phone, and option fields.
Time	Natural language date/time parsing. Handles relative offsets, weekday targets, timezone conversions, and complex expressions like "next Friday at 3pm EST".
Code Search	Semantic code search across indexed repositories using vector embeddings (ChromaDB). Search by natural language query, get back relevant code snippets with file paths and line numbers.

---

Built-in MCP Servers

These are MCP servers that Synapse bundles and manages. Enabled automatically when configured.

Server	What It Does
Filesystem (@modelcontextprotocol/server-filesystem)	Full read/write access to your local code repositories. Configure which paths to expose in Settings → Repos.
Google Workspace (workspace-mcp)	Gmail (read, search, send), Google Drive (list, read, create files), and Google Calendar (events, scheduling). One-click OAuth setup in Settings.
Playwright (@playwright/mcp)	Headless browser control — available separately from the native Browser tool for automation without UI.
Sequential Thinking (@modelcontextprotocol/server-sequential-thinking)	Structured step-by-step reasoning for complex, multi-stage problems. Agents break tasks into explicit thought chains before acting. Enabled by default.
Memory (@modelcontextprotocol/server-memory)	Persistent knowledge graph memory across sessions. Agents store and retrieve facts, relationships, and context between runs. Enabled by default.

---

Sandbox Tool Details

The Python sandbox is one of the most powerful tools. It runs code in a Docker container with:

• Libraries pre-installed: pandas, numpy, matplotlib, scikit-learn, requests, beautifulsoup4, openpyxl, and more
• Vault access: read and write files to the persistent vault directly from Python
• Resource limits: 512 MB RAM, 1 CPU — protects the host machine
• Isolation: each execution runs in a fresh container context

Example use cases:

• Data analysis and visualization
• File processing (CSV, Excel, JSON)
• API calls and data transformation
• Statistical modeling
• Generating charts saved to vault

---

Code Search Tool Details

The Code Search tool indexes your repositories and enables semantic search:

1. Link a repository in Settings → Repos
2. Click Index Repository — Synapse builds a ChromaDB vector index
3. Agents can now search with natural language: "find where API keys are validated" → returns file paths and line numbers

Useful for large codebases where exact keyword search falls short.

---

Tool Naming in Agents

When you restrict an agent's tools, use the tool function names. Key tool names:

Tool Function	Tool Server
execute_python	Sandbox
vault_write, vault_read, vault_list, vault_patch	Vault
list_tables, get_table_schema, run_sql_query	SQL Agent
browser_navigate, browser_snapshot, browser_click	Browser
parse_pdf	PDF Parser
parse_xlsx	Excel Parser
scrape_url, crawl_urls	Web Scraper
collect_data_form	Collect Data
get_current_time, parse_datetime	Time
search_code	Code Search

---

See Also

• MCP Servers — connect external MCP servers
• Custom Tools — add your own tools via webhooks or Python scripts
• Vault — persistent file storage details