About Odichat
Odichat is an MCP server that makes the Odichat API available to AI assistants for managing customer communications and related operations. Its 123 tools cover core inbox functions as well as Odichat-specific Kanban boards and scheduled messages.
Key features
- 01123 tools covering Odichat API endpoints
- 02Manage contacts, conversations, and messages
- 03Work with teams, inboxes, and agents
- 04Access reports and help center functions
- 05Manage Kanban boards, steps, and tasks
- 06Create and manage scheduled messages
Use cases
- 01Manage customer conversations from an AI assistant
- 02Update contacts and assign conversations
- 03Review reports on agents, inboxes, and teams
- 04Organize work with Kanban tasks and scheduled messages
MCP support
Exposes Odichat API operations as MCP tools over stdio or stateless Streamable HTTP. In HTTP mode, each request supplies its Odichat API token.
Configuration
Remote servers need no local install. Pick your client, then copy the config or install in one click.
https://mcp.odichat.app/mcp?token={token}claude mcp add --transport http odichat-mcp-odichat https://mcp.odichat.app/mcp?token={token}{"mcpServers":{"odichat-mcp-odichat":{"type":"http","url":"https://mcp.odichat.app/mcp?token={token}"}}}Open in Cursor{"servers":{"odichat-mcp-odichat":{"type":"http","url":"https://mcp.odichat.app/mcp?token={token}"}}}Open in VS Code{"mcpServers":{"odichat-mcp-odichat":{"type":"http","url":"https://mcp.odichat.app/mcp?token={token}"}}}{"mcpServers":{"odichat-mcp-odichat":{"serverUrl":"https://mcp.odichat.app/mcp?token={token}"}}}Run commands in a terminal; paste JSON into the client’s MCP config and restart. Servers that need sign-in will prompt on first connect.
Tools
Details
From the project READMEmcp-odichat
A Model Context Protocol (MCP) server that exposes the full Odichat API — including Odichat exclusive features — as 123 tools for use with AI assistants like Claude, ChatGPT, VS Code Copilot, and others.
Features
- 123 tools covering all Odichat API endpoints
- Account, Agents, Contacts, Conversations, Messages, Inboxes, Teams, and more
- Reports (v1 & v2), Help Center, Automation Rules, Custom Attributes, Custom Filters
- Odichat exclusive: Kanban Boards, Kanban Steps, Kanban Tasks, Kanban Audit Events, Scheduled Messages
- Proper MCP tool annotations (
readOnlyHint,destructiveHint,idempotentHint) - Multi-account support (
account_idis a per-tool argument) - Two transports:
- stdio — local subprocess (Claude Desktop, VS Code, and any MCP client)
- Streamable HTTP — remote, hosted server (Claude Web/Desktop/Code, ChatGPT). Stateless; the API token is passed per request via a
?token=query param
Requirements
- Bun v1.0+
- An Odichat instance with API access
Installation
bun install
Environment Variables
| Variable | Required | Description |
|---|---|---|
ODICHAT_API_TOKEN | stdio only | API access token (found in Odichat → Profile → Access Token). Not used by the HTTP transport. |
MCP_TRANSPORT | No | Set to http to run as a network server. Any other value (or unset) uses stdio. |
PORT | No | Port for the HTTP transport. Defaults to 3000. |
In HTTP mode the token is not read from the environment — each client supplies its own token via the
?token=query param, keeping the server stateless and multi-tenant.
Usage
Development
bun run dev
Production (stdio)
bun run start
Production (remote HTTP server)
Run the server over HTTP with Bun.serve() and the web-standard Streamable HTTP transport:
MCP_TRANSPORT=http PORT=3000 bun run src/index.ts
Or with Docker:
docker build -t mcp-odichat .
docker run -p 3000:3000 mcp-odichat
The server is stateless — no database, no sessions. Each request must include the
caller's Odichat API token as a ?token= query param. SSL is expected to be
terminated by an upstream proxy (e.g. Traefik).
Endpoints:
| Method | Path | Description |
|---|---|---|
POST/GET | /mcp | MCP Streamable HTTP endpoint |
GET | /health | Health check (returns ok) |
Connecting to a hosted server
Point any remote-MCP-capable client (Claude Web/Desktop/Code, ChatGPT) at the server URL with your token in the query string:
https://mcp.odichat.app/mcp?token=YOUR_API_TOKEN
- In Odichat, go to Profile → Access Token and copy your token.
- Add the URL above (with your token) as the MCP server URL in your AI client.
Local (stdio) client configuration
Claude Desktop Configuration
Add to your claude_desktop_config.json:
{
"mcpServers": {
"odichat": {
"command": "bun",
"args": ["run", "/path/to/mcp-odichat/src/index.ts"],
"env": {
"ODICHAT_API_TOKEN": "your-api-token"
}
}
}
}
VS Code Configuration
Add to your .vscode/mcp.json:
{
"servers": {
"odichat": {
"command": "bun",
"args": ["run", "${workspaceFolder}/src/index.ts"],
"env": {
"ODICHAT_API_TOKEN": "your-api-token"
}
}
}
}
ChatGPT Configuration
ChatGPT connects to remote MCP servers over HTTP — see Connecting to a hosted server above.
Available Tools (123)
Account (2)
account_get, account_update
Agent Bots (5)
agent_bots_list, agent_bots_create, agent_bots_get, agent_bots_update, agent_bots_delete
Agents (4)
agents_list, agents_create, agents_update, agents_delete
Audit Logs (1)
audit_logs_list
Automation Rules (5)
automation_rules_list, automation_rules_create, automation_rules_get, automation_rules_update, automation_rules_delete
Canned Responses (4)
canned_responses_list, canned_responses_create, canned_responses_update, canned_responses_delete
Contacts (11)
contacts_list, contacts_create, contacts_get, contacts_update, contacts_delete, contacts_conversations, contacts_search, contacts_filter, contacts_create_contact_inbox, contacts_contactable_inboxes, contacts_merge
Contact Labels (2)
contact_labels_list, contact_labels_set
Conversations (12)
conversations_meta, conversations_list, conversations_create, conversations_filter, conversations_get, conversations_update, conversations_toggle_status, conversations_toggle_priority, conversations_set_custom_attributes, conversations_get_labels, conversations_set_labels, conversations_reporting_events
Conversation Assignments (1)
conversation_assignments_assign
Messages (3)
messages_list, messages_create, messages_delete
Custom Attributes (5)
custom_attributes_list, custom_attributes_create, custom_attributes_get, custom_attributes_update, custom_attributes_delete
Custom Filters (5)
custom_filters_list, custom_filters_create, custom_filters_get, custom_filters_update, custom_filters_delete
Help Center (5)
help_center_portals_list, help_center_portals_create, help_center_portals_update, help_center_categories_create, help_center_articles_create
Inboxes (10)
inboxes_list, inboxes_get, inboxes_create, inboxes_update, inboxes_get_agent_bot, inboxes_set_agent_bot, inbox_members_list, inbox_members_create, inbox_members_update, inbox_members_delete
Integrations (4)
integrations_list_apps, integrations_create_hook, integrations_update_hook, integrations_delete_hook
Profile (1)
profile_get
Reports (9)
reports_account_overview, reports_account_summary, reports_agent_summary, reports_conversation_metrics, reports_v2_overview, reports_v2_agents, reports_v2_inboxes, reports_v2_teams, reports_v2_labels
Teams (9)
teams_list, teams_create, teams_get, teams_update, teams_delete, team_members_list, team_members_add, team_members_update, team_members_delete
Webhooks (4)
webhooks_list, webhooks_create, webhooks_update, webhooks_delete
Kanban Boards (6) — [Odichat]
kanban_boards_list, kanban_boards_create, kanban_boards_get, kanban_boards_update, kanban_boards_delete, kanban_boards_update_agents
Kanban Steps (5) — [Odichat]
kanban_steps_list, kanban_steps_create, kanban_steps_get, kanban_steps_update, kanban_steps_delete
Kanban Tasks (6) — [Odichat]
kanban_tasks_list, kanban_tasks_create, kanban_tasks_get, kanban_tasks_update, kanban_tasks_delete, kanban_tasks_move
Kanban Audit Events (2) — [Odichat]
kanban_audit_events_list, kanban_audit_events_get
Scheduled Messages (4) — [Odichat]
scheduled_messages_list, scheduled_messages_create, scheduled_messages_update, scheduled_messages_delete
Development
# Run tests
bun test
# Lint & format
bun run lint
bun run format
# Type check
bun run build-check
# Full check (lint + type-check + tests)
bun run check
License
MIT
Comments
New to MCP?
From what the protocol is to installing your first server, these answers cover most first questions.
An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.