Everything you need to give your AI assistant private structured memory for real-world records: how schema generation works, how to connect each major assistant, and how access control, tokens, and the OpenAPI contract fit together.
ChatData Sync gives your AI assistant a private, structured place to read and write real-world records. You describe what you want to track in plain English, ChatData Sync generates the data model and assistant actions, and you connect your assistant with a single copy-paste step.
This is especially useful for solo service operators and side hustlers who already use ChatGPT or Claude daily and need reliable recall across chats.
For the full conceptual overview, see How it works.
A good description names the things you want to track and how they relate. You do not need technical vocabulary — entity names, fields, and relationships are inferred.
You can type or speak your description: the setup wizard includes a microphone button that transcribes what you say (powered by OpenAI Whisper) straight into the description box. A typed or spoken description is always required before you import a CSV/Excel file or build a tracker by hand — it is what teaches the generator and your assistant what the tracker is for.
Example:
Track customers and their vehicles. Each vehicle can have
multiple repair orders. Each repair order includes inspection
findings, parts used, an estimate with line items, customer
approval, and technician notes. Track open vs. completed
orders and which invoices are unpaid.Schema descriptions are normalized and scanned for prompt-injection patterns before being sent to the schema generator. See Security for details on input hardening.
This auto-repair example is a practical proof point: it shows how one assistant can reliably capture business operations, not just generic notes.
Before any data is created, you see the full structure — tables, fields, types, and relationships. You can adjust field names, remove tables you do not need, or approve as-is.
Once approved, the tracking space is live immediately. The data model, role-based access policy, assistant action endpoints, and OpenAPI contract are all generated together.
Editing after creation: a live tracker is never locked. From the tracker's layout editor you can add new sections (tables) and fields up to your plan limits, change a field's type, and create a link field that relates one section to another (for example, a Payments section that links to a Clients section). Your existing records stay safe. The same additive edits are also available to your assistant through the MCP tools below, so you can grow a tracker by conversation.
Each tracking space generates two connection options. Pick the one your assistant supports.
Paste a single connection string into a custom GPT's Actions settings. Ready in under two minutes.
Register your tracker's MCP endpoint with a scoped token in Claude.ai or the Anthropic API.
Point any Model Context Protocol client at https://chatdatasync.com/api/mcp over streamable HTTP. See the MCP server reference below.
The same contract works for any client that consumes OpenAPI 3.1 specs and bearer tokens.
Every tracking space exposes an OpenAPI 3.1 specification that mirrors the generated data model. The contract is what assistants use to discover available actions: create a record, list records, fetch by ID, patch, and delete.
The reference spec for the public actions surface is available at /openapi.actions.json. Your tracking space provides a project-specific contract under /api/v1/projects/{projectId}/openapi.json.
ChatData Sync runs a Model Context Protocol (MCP) server so MCP-compatible assistants and agents — Claude, ChatGPT, and any custom MCP client — can read and write your private tracker through conversation. The same server backs every tracking space.
https://chatdatasync.com/api/mcp (POST for JSON-RPC calls; GET returns a capability summary).2025-06-18.x-api-key header. Best for direct, single-tracker setups; no login flow required.S256), and Dynamic Client Registration so connectors configure themselves.OAuth discovery endpoints (clients find these automatically — you do not need to copy them):
https://chatdatasync.com/.well-known/oauth-protected-resource/api/mcphttps://chatdatasync.com/.well-known/oauth-authorization-serverhttps://chatdatasync.com/oauth/registermcp:read, mcp:write, offline_accessTools the server exposes: get_sync_context, list_records, search_records, summarize_records, get_record, create_record, update_record, update_records (bulk), and delete_record for working with records, plus list_trackers and create_tracker for managing trackers. Layout tools (owner-only, additive): describe_schema returns the editable layout and remaining capacity, add_section adds a new section (table), and add_field adds a field — including choice fields and reference (link) fields that relate one section to another. These never delete existing data. Resources: chatdatasync://project/current/context and chatdatasync://project/current/schema.
The scoped key and OAuth tokens both authorize a single tracker only — they cannot reach your account or any other tracker. For a step-by-step walkthrough, see Connect Claude via MCP.
Account-level operations (creating trackers, schema generation) require a Firebase ID token. Assistant access uses a project-scoped API key passed as x-api-key. The key authorizes sync operations on one specific tracker only — it cannot reach your account or any other tracker.
Keys can be rotated from the tracker's settings. Rotation is rate-limited and requires the existing key. Once rotated, the previous key is invalidated immediately and any assistant configured with it must be reconnected.
Each tracker also has a hard record cap (Free 500, Single 5,000, Side Hustle 15,000, Growing Business 30,000, Team 50,000). Add-ons raise these numbers without changing your plan: Record Boost (+10,000 records per tracker, $5/month, stackable up to the 50,000-per-tracker maximum), Action top-up (+500 assistant actions, $5 one-time), and Action add-on (+1,000 assistant actions/month, $7/month).
Single Tracker is the easiest paid starting point when you want one focused workflow. Bundle plans lower your effective per-tracker cost as you grow. All plans include CSV export. See Pricing for the full breakdown.
Request guardrails are enforced on every endpoint: MAX_BODY_BYTES=262144 (256 KiB), MAX_OBJECT_DEPTH=8, MAX_OBJECT_FIELDS=4000, and MAX_TEXT_LENGTH=4000 per string. Endpoint-specific rate limits apply to AI, schema, sync, and key rotation routes.
On 429 Too Many Requests, read X-RateLimit-Reset (unix seconds) and retry after reset. Recommended fallback is exponential backoff with jitter (for attempt n: delayMs = min(30000, 500 * 2^n) + random(0..250)).
The full conceptual picture: problem, three-step setup, and what gets generated.
Tenant isolation, scoped tokens, default-deny Firestore rules, and prompt sanitization.
Real workflows including auto repair, service businesses, and assistant integrations.
Common questions about setup, pricing, supported assistants, and data handling.
The longer-form explanation of why structured assistant data matters and how ChatData Sync delivers it.
How ChatData Sync compares to spreadsheet-style tools for AI-assistant workflows.
Describe your workflow in plain English. Review the generated schema. Connect your assistant. The Free plan includes one tracking space and 200 assistant actions per month.
Start free