Overview

The Fibery MCP Server connects your Fibery workspace to any LLM client that implements the Model Context Protocol (MCP). Once installed, the server exposes Fibery data and actions (list databases, describe schema, query, create/update entities) over MCP so an MCP-compatible LLM (for example, Claude for Desktop) can access and manipulate your workspace using natural language.

This is useful when you want conversational AI assistants to read your workspace structure and records, answer questions about projects, or create and modify entities without writing custom integration code. The server acts as a thin bridge that authenticates to Fibery using your API token and translates MCP tool calls into Fibery API requests.

Features

Query Fibery entities using natural language through an MCP client
Inspect workspace structure: databases, fields, and types
Create single or batch entities with specified field values
Update existing entities programmatically
Works with any MCP-compatible LLM client (e.g., Claude Desktop)
Small, open-source server (MIT license) ready for local or development use

Installation / Configuration

Prerequisites:

Fibery account and a Fibery API token
Python 3.10+
Either Smithery (for automated installs) or uv (for local tool management)

Install via Smithery (automatic for Claude Desktop)

npx -y @smithery/cli install @Fibery-inc/fibery-mcp-server --client claude

Install via uv

# Install the tool
uv tool install fibery-mcp-server

Run the tool with Fibery credentials:

uv tool run fibery-mcp-server \
  --fibery-host your-domain.fibery.io \
  --fibery-api-token your-api-token

If the uv command is not on your PATH, use its absolute path:

/Users/<you>/.local/bin/uv tool run fibery-mcp-server --fibery-host ... --fibery-api-token ...

Development (run from cloned repo)

uv --directory path/to/cloned/fibery-mcp-server run fibery-mcp-server \
  --fibery-host your-domain.fibery.io \
  --fibery-api-token your-api-token

Example MCP client configuration (Claude Desktop: Settings → Developer → Edit Config). Add an MCP server entry to run the uv command:

{
  "mcpServers": {
    "fibery-mcp-server": {
      "command": "uv",
      "args": [
        "tool",
        "run",
        "fibery-mcp-server",
        "--fibery-host",
        "your-domain.fibery.io",
        "--fibery-api-token",
        "your-api-token"
      ]
    }
  }
}

Available Tools

The MCP server exposes a set of tools (MCP endpoints) that LLM clients can call.

Tool name	Purpose	Key parameters
list_databases	Return all databases (entity types) in the workspace	none
describe_database	Show schema for a given database (fields, types)	database identifier (name/ID)
query_database	Run queries against a database using Fibery API filters	database, filter/query, limit, fields
create_entity	Create a single entity with specified fields	database, field-value map
create_entities_batch	Create multiple entities in one request	database, array of field-value maps
update_entity	Update fields for an existing entity	entity ID, field-value map

Example: create_entity request payload (conceptual)

{
  "database": "Tasks",
  "fields": {
    "Name": "Write design doc",
    "Assignee": "[email protected]",
    "Status": "In Progress"
  }
}

Use Cases

Natural-language queries: Ask the LLM “Show me all high-priority bugs assigned to Bob” and have it call query_database, returning a digestible summary of matching entities.
Schema-aware assistance: Use describe_database to let a model understand available fields and types before suggesting or filling values in conversational form.
Create new records from conversation: In a chat with an assistant, say “Create a new feature request titled ‘Export CSV’ assigned to Product”, and the assistant will call create_entity with correct field mappings.
Bulk imports via conversation: Provide several tasks in a single message and let the assistant call create_entities_batch to create them all at once.
Safe updates: Ask the assistant to update status or assignee for a specific entity; it will use update_entity with an explicit entity ID and provided fields.

Tips and Notes

Keep your Fibery API token secure. Do not check tokens into source control.
Use describe_database before creating/updating entities so the LLM can map human-friendly names to Fibery field identifiers and types.
The server is designed to be minimal—most client integration logic (how to map conversational intents to tool calls) lives on the MCP client/LLM side.
Repository and source code: https://github.com/Fibery-inc/fibery-mcp-server

If you need to extend behavior, clone the repo and run via uv with –directory for local development. The project is licensed under MIT, allowing modifications for internal workflows.

Fibery MCP Server for LLM Workspace Integration