Skip to main content
Glama

Kintio

Connect chat channels to an Agent you control.

English | 简体中文

CI CodeQL Latest release License

Deploy Kintio next to an Agent runtime and authorized people can assign work, steer an active task, and receive results without leaving their chat app. Kintio connects each conversation to the Agent while keeping identities, state, and reply capabilities scoped to the originating channel. The current runtime uses the Codex CLI session on the deployment host.

Chat: "Investigate this"  -> Kintio starts a task in the Agent on your machine
Chat: "Focus on the logs" -> Kintio steers that running task
Agent result              -> Kintio delivers it back to the same chat

What it provides

  • A separate Agent thread for every adapter identity, with no implicit identity or history merging across channels.

  • Steering for active work: a user's follow-up can redirect the current Agent turn instead of waiting behind stale instructions.

  • A durable SQLite inbox, restart recovery, and live-message priority over backlog recovery.

  • Session-scoped MCP actions that cannot choose another recipient or expose provider credentials, raw user identifiers, or database paths to the Agent.

  • Explicit delivery outcomes: accepted, failed, or uncertain. Uncertain sends are not blindly retried.

One deployment currently shares one Agent runtime login. Conversation state is isolated, but Kintio is not yet a multi-tenant platform where each chat user supplies independent Agent credentials, quotas, or working directories.

Related MCP server: Agent Communication MCP Server

Current adapters

Adapter

Inbound

Outbound

Transport

Conversation identity

WeChat KF API

Text and images; other message types become explicit summaries

Text, images, links, Mini Programs, and locations

Public HTTPS callback

open_kfid + external_userid

Weixin iLink

Text and images; other message types become explicit summaries

Text and images

Long polling after QR-code binding

bot_id + user_id

Adapter capabilities are enforced per conversation. For example, an image received in one conversation can only be referenced by the short-lived MCP session bound to that same conversation.

Quick start

Prerequisites:

  • Node.js 24 or later;

  • pnpm 10 (pinned in package.json);

  • an installed and authenticated Codex CLI;

  • credentials for at least one supported adapter.

corepack enable pnpm
pnpm install --frozen-lockfile
cp .env.example .env
codex login status

No adapter is enabled by default. Follow the setup guide to generate a strong MCP token and configure one adapter:

  • For WeChat KF API, set its callback token, EncodingAESKey, CorpID, and secret. A temporary WECOM_AUTH_TRIGGER can authorize the first user without knowing their external_userid in advance.

  • For an existing Weixin iLink binding, set ILINK_ENABLED=true. Creating a new binding currently starts from an authorized WeChat KF conversation.

Start Kintio:

pnpm start

Successful startup prints Hono server is listening on port 8888. Complete the callback or binding checks described in the setup guide before sending production traffic.

Security boundaries

  • .env, SQLite databases, downloaded media, and local key files are ignored by Git and must never be force-added.

  • The Agent does not receive provider secrets, stable user identifiers, raw media IDs, or database paths.

  • MCP endpoints require strong bearer tokens; non-loopback access must use HTTPS.

  • Project-level Agent capability restrictions are not an operating-system sandbox. Use a dedicated system account and additional isolation appropriate to the Agent's real powers.

  • A provider accepting an outbound request does not prove that a client displayed it; uncertain outcomes remain explicit to avoid duplicate delivery.

See the security policy for the complete trust boundary and private vulnerability-reporting process.

Documentation and contributing

The default verification entry point is:

pnpm test

Tests use temporary SQLite databases and simulated provider and Agent boundaries by default. They contact real services only when explicit live-test flags, targets, and allowlists are supplied.

Project status

Kintio is in the 0.x stage. It currently ships the two adapters listed above while the channel, identity, recovery, and Agent-runtime contracts are stabilized for future adapters.

The Weixin iLink protocol implementation incorporates work described in THIRD_PARTY_NOTICES.

Kintio is an independent open-source project. It is not affiliated with, authorized by, endorsed by, or an official product of Tencent, WeChat, Weixin, or any other channel provider.

A
license - permissive license
Not graded
quality - not tested
A
maintenance

Maintenance

Maintainers
9hResponse time
Release cycle
1Releases (12mo)
Commit activity

Resources

Unclaimed servers have limited discoverability.

Looking for Admin?

If you are the server author, to access and configure the admin panel.

Related MCP Servers

View all related MCP servers

Related MCP Connectors

  • Agent-to-agent network for teams: dm, who-knows-X routing, shared rooms. Human-in-the-loop.

  • Create and manage AI agents that collaborate and solve problems through natural language interacti…

  • Ephemeral REST chatrooms for AI agents to coordinate. Share a room URL — agents talk live.

View all MCP Connectors

Latest Blog Posts

MCP directory API

We provide all the information about MCP servers via our MCP API.

curl -X GET 'https://glama.ai/api/mcp/v1/servers/Gkxie/kintio'

If you have feedback or need assistance with the MCP directory API, please join our Discord server