Syntitan MCP

What syntitan-mcp MCP server does

The syntitan-mcp MCP server exposes Syntitan’s data-management capabilities to MCP-compatible clients. It is aimed at examining datasets before AI use, identifying readiness issues, inspecting structural and column-level details, and applying supported preprocessing or de-identification steps.

Read-only operations include listing accessible datasets, retrieving schemas, profiling columns, reviewing version history, checking refinement options and status, and examining AI Readiness. AI Readiness results include an overall score, category-level scores such as Privacy and Conciseness, and priority issues. Category details can identify the columns that contributed to a result and the size of their contribution.

Sampling is privacy-aware: dataset samples mask PII columns, and requests are refused when unmasked PII is present. Refinement is the write operation. It consumes credits, runs preprocessing or de-identification, and saves the result as a new dataset snapshot.

How it works

The syntitan-mcp MCP server is hosted remotely and uses Streamable HTTP at https://mcp.syntitan.ai/mcp. A client sends MCP requests to that endpoint, while Syntitan handles access to the user’s datasets and returns the requested result.

Authentication uses OAuth 2.0. On the first connection, the client opens Syntitan’s sign-in page in a browser. After the user signs in and grants access, the client connects without storing an API key. Communication is served over HTTPS/TLS.

The server also provides a diagnosis guide for interpreting AI Readiness metrics and scores. Before starting a refinement, clients can request an estimate of the credits and time involved, then check progress through the refinement-status tool.

Setup and configuration

Add the remote MCP server in a client that supports remote MCP connections, using this endpoint:

https://mcp.syntitan.ai/mcp

No local process or environment variable is documented for this hosted connection. Complete the browser-based Syntitan authorization when prompted. The README specifically identifies Cursor, VS Code, Claude Code, and other MCP clients as supported connection options.

Tools and capabilities

Available operations cover:

• Listing accessible datasets with latest version, dimensions, AI Readiness score, and modification time.
• Returning dataset schemas without exposing PII values.
• Profiling a column with missing-rate, distinct-value, top-value, and example statistics.
• Returning AI Readiness summaries and category-level explanations.
• Retrieving dataset samples with automatic PII masking.
• Reviewing dataset versions, authors, release states, timestamps, and recorded changes.
• Checking available preprocessing and de-identification choices.
• Estimating refinement cost and duration, starting a refinement, and checking its status.

Limitations and notes

Refinement is not read-only and uses credits. It produces a new snapshot rather than silently replacing the source version. Sample requests are restricted when unmasked PII remains in the dataset, and PII columns are only returned in masked form through that sampling capability.

Syntitan states that it accesses data required for the user’s request. Audit logs may contain identifiers such as email addresses and are retained for operational and security purposes under its privacy policy. Data retention, storage, collection, and deletion details are governed by Syntitan’s published privacy policy and terms of service.