Manage your Gnosaris from any AI chat
What the MCP server does
The Model Context Protocol is the open standard AI clients use to call external tools. Gnosari runs one MCP server at a single address, so the client you already have open can create a Gnosari, attach knowledge to it, publish it at a joina.chat link and read the structured data it collects, all through 16 tools and ordinary conversation. Authentication is OAuth, handled by the client: there is no API key to paste and no config file for most clients.
Connect in seconds
https://mcp.gnosari.com{
"mcpServers": {
"gnosari": {
"transport": "streamableHttp",
"url": "https://mcp.gnosari.com"
}
}
}Clients this works in
Every client below speaks MCP over HTTP. The ones marked for a config file take the JSON above; the rest take the URL on its own.
What a build looks like
01 · You ask
“Build me a Gnosari for my dental practice that books first visits. Read oaklinedental.com, collect name, phone, insurance and whether they are nervous, then give me the link.”
02 · It calls the tools
- Create GnosariName, purpose, the fields to collect
- Manage KnowledgePoints it at your website
- Manage TraitsAttaches how it should behave
- Manage Agent LinksPublishes a link and a QR
03 · It is live
Published
joina.chat/l/first-visit
A real, published Gnosari. It shows up in the app like any other, and the records land in the same inbox.
Add this server
https://mcp.gnosari.com
Paste it into your client's connector settings. Sign in to Gnosari when it asks. There is no key to copy and nothing to install.
Works in
- Claude
- ChatGPT
- Claude Code
- Cursor
- Cline
- Windsurf
The whole thing is a conversation
What you can do from the chat
- Composite Agent CreationCreate a fully configured, data-collecting agent in one call. gnosari_create atomically creates the agent, data collection template, and optional publish - instructions, welcome screen, and data fields all required up front.
- Conversation Data CollectionYour agents automatically extract structured data from every conversation. View records, filter by status and template, and get dashboard statistics - all from your AI chat.
- Knowledge and ConfigurationAttach website knowledge for RAG, assign personality traits, and configure access levels. Each resource tool covers full CRUD plus assign/remove via a single action parameter.
Full tool reference
| Category | Tool | What it does |
|---|---|---|
| Agent Lifecycle | gnosari_create | Create a fully configured, data-collecting AI agent in one atomic call. Requires name, instructions, welcome screen fields, and a data_collection template. Optionally publishes immediately |
| Agent Lifecycle | gnosari_get | Get a complete overview of an agent - identity, instructions, access, appearance, knowledge, traits, data collection, and readiness block |
| Agent Lifecycle | gnosari_update | Update agent identity fields (name, description, model, temperature) |
| Agent Lifecycle | gnosari_delete | Two-phase delete with confirmation. First call returns a warning; second call with confirmed=True permanently deletes |
| Agent Configuration | gnosari_manage_instructions | Replace, append, or prepend agent instructions. Use action param to control how the update is applied |
| Agent Configuration | gnosari_manage_access | Set access level (PUBLIC, PRIVATE, PASSWORD_PROTECTED) and public URI. Publishing domain resolved server-side |
| Agent Configuration | gnosari_manage_appearance | Configure greeting, empty state, suggested prompts, and chat theme |
| Resource Management | gnosari_manage_traits | Create, list, update, delete personality traits and assign/remove them from agents. Pass gnosari_id to create and assign in one call |
| Resource Management | gnosari_manage_knowledge | Create, list, update, delete URL-based knowledge sources and assign/remove them from agents. Sources are crawled and indexed for RAG |
| Resource Management | gnosari_manage_data_collection | Create, list, update, delete data collection templates and assign/remove them from agents. Supports collection_mode and custom_prompt for guided flows |
| Discovery and Data | gnosari_search | Unified search across agents, traits, templates, and knowledge sources by name |
| Discovery and Data | gnosari_check_uri | Check if a URI slug is available before publishing. Returns availability and hints with alternatives if taken |
| Discovery and Data | gnosari_collected_data | Browse data extracted from conversations (action="list") or get dashboard statistics (action="stats"). Filter by agent, template, status, and date range |
| Discovery and Data | gnosari_manage_link | Create, list, update, and delete shareable agent links with chat or conversation presentation. Every link ships with a QR image and supports remapping without reprinting |
| Utility | gnosari_embed_code | Generate the HTML snippet to embed the Gnosari chat widget on any website |
| Utility | gnosari_health | Check MCP server health and connectivity |
Frequently Asked Questions
Everything you need to know about getting started. Can't find the answer you're looking for? Reach out to our team.
What is the Gnosari MCP server?
The Gnosari MCP server lets an AI assistant such as Claude, ChatGPT or Cursor create and manage your Gnosari agents through the Model Context Protocol, an open standard for connecting assistants to tools and data. From your own AI chat you build agents, set what they collect, read collected records and configure knowledge sources, with the same result as the web dashboard.
Which AI tools support MCP?
The Gnosari MCP server works with Claude Desktop and claude.ai (native Connectors support), ChatGPT (Plus and Pro), Claude Code (CLI), Cursor, Cline (VS Code) and Windsurf. Most clients support URL-paste setup with automatic OAuth, so no config files are needed: paste the server address and sign in.
How do I connect my AI client?
Copy the MCP server URL (https://mcp.gnosari.com) and paste it into the MCP settings of your AI client. For Claude Desktop, go to Settings, then Connectors, then Add custom connector. For ChatGPT, go to Settings, then Apps and Connectors. Your client handles authentication automatically - just sign in to Gnosari when prompted.
Is my data secure?
Yes. Gnosari uses OAuth 2.1 with PKCE for authentication - the same standard used by Google, GitHub, and other major platforms. All connections use HTTPS/TLS, tokens are short-lived and auto-refreshed, and you can revoke access anytime. API keys are also available as an alternative for automation workflows.
What can I do with the MCP server?
With the Gnosari MCP server you create and configure AI agents with knowledge sources and data collection fields, view collected data and statistics, check URI availability before publishing, generate embed code and manage QR links, all through natural conversation with your AI assistant. A marketing manager builds a lead-capture agent in one chat; a developer scripts the same through the API.
How does data collection work via MCP?
When you create a Gnosari agent with gnosari_create, a data collection template is configured as part of the same atomic call. Your agents then automatically extract structured data from conversations. Use gnosari_collected_data with action="list" or action="stats" to review what your agents have gathered.
What are joina.chat links?
joina.chat is the public domain where Gnosari publishes your AI agents. When you create an agent with publish=True, it gets a public URL like joina.chat/your-uri where anyone can start a conversation. Use gnosari_manage_access to control who can access your agent.
Can I use both the web UI and MCP?
Yes. Changes made through the Gnosari MCP server appear instantly in the web dashboard and the other way round, because both write to the same account. Use whichever fits the moment: many people set agents up quickly from their AI assistant and use the dashboard for visual configuration and reading records.
Is there rate limiting?
Yes. The Gnosari free tier allows 100 MCP operations per hour, and paid plans have higher limits based on the subscription. A business with enterprise volume can contact us for custom rate limits. Reads and writes count the same, and the limit is per account, not per assistant.