Skip to main content
Glama

Teleloom connects your Telegram accounts and bots to the agent you already use. Find messages across chats, catch up on discussions, build summaries with source links, inspect documents, and prepare actions you can review before they run.

One local owner manages connections and state. Multiple agent clients share it; you choose the accessible chats and the permissions for each account.

What you can do

Task

What Teleloom provides

Catch up

Read history, search messages and authors, follow replies and topics, review unread messages

Research many chats

Bounded collection jobs, frozen originals, counters, excerpts, exports and explicit coverage gaps

Write a verifiable digest

Stable evidence references, source links, claim/revision validation and portable digest instructions

Work with media

Photos, rich text, documents/PDF, optional OCR and opt-in transcription

Act with control

Preview and confirm messages, replies, media, bounded broadcasts, and supported account/group changes

Keep context manageable

Reusable field selection, compact views, output budgets and access to the exact originals

67 MCP tools and six portable skills: connect, read, inbox, digest, send and broadcast. See the tool reference for exact contracts and limits.

Related MCP server: telegram-mcp

Quick start

Requires Git, uv and Python 3.12–3.14. User accounts need your own Telegram API ID/hash from my.telegram.org. Bots use a bot token. Credentials are entered locally, outside an agent conversation.

git clone --single-branch https://github.com/starsinc1708/teleloom.git
cd teleloom
uv sync --frozen
uv run teleloom init
uv run teleloom auth user --profile personal --ui browser
uv run teleloom profile list

On your phone, open Telegram → Settings → Devices → Link Desktop Device and scan the QR. A working OS credential store is required to save the session. For headless hosts, bots, read restrictions and first-use troubleshooting, follow the English guide or Russian guide.

Connect a client and install its skills:

uv run teleloom config client --client codex
uv run teleloom skills install --client codex

Merge the printed fragment into the existing client configuration, then reconnect MCP. Use claude, opencode (v2), opencode-v1, hermes or pi for other clients. Existing skills are preserved unless you explicitly request replacement. Client paths and discovery checks.

For OpenCode 2 or Hermes, register MCP through the installed client's CLI:

uv run teleloom config client --client opencode --install
uv run teleloom skills install --client opencode
uv run teleloom config client --client hermes --install
uv run teleloom skills install --client hermes

Choose the pair for your client, then reconnect MCP (opencode reload for a running OpenCode service). --install is available in the current source checkout; v0.5.0 only prints fragments. An identical connection is preserved; a conflicting teleloom entry requires manual review. Hermes skills follow hermes config path, including its active profile and HERMES_HOME.

Try this first:

Check server status and list profiles. Ask me to select a profile and one chat. Read its latest 20 messages without marking them as read. Summarize the main topics with message links and state any missing context. Do not send messages or change settings or permissions.

Control and privacy

  • New profiles can read all chats accessible to the account. Use selected read policy to restrict access; the setup guide explains the independent tool and chat gates.

  • Sending and other external changes require their own permissions and an immutable preview plus confirmation. An uncertain delivery is recorded as unknown and is never automatically resent.

  • An explicit human instruction to send to exact recipients can authorize that previewed send and selected local files through owner_authorized=true. Permanent allowlists and file roots stay unchanged. Plan authorization.

  • Telegram credentials stay in the OS keyring or supplied environment. State and exports stay in your local data directory. Your chosen agent/model provider receives the evidence you request; optional external analysis is explicit.

  • Reading does not acknowledge messages. Bot history covers collected observations, and search/digest results disclose their scope and gaps.

The confirmation flag is trusted input from your agent client; it is not independent proof of a human approval. All local clients share one trusted OS owner. Security model and vulnerability reporting.

Clients and optional features

Client

Integration

Codex

Stdio MCP and six portable skills

Claude Code

Stdio MCP and six portable skills

OpenCode 2 / 1

Native MCP configuration; separate version targets

Hermes

Native mcp_servers configuration and skills

Pi

Native mcpServers configuration and skills

Claude Desktop

MCP configuration; GUI behavior remains unverified

Other MCP hosts

Stdio or authenticated loopback Streamable HTTP

Optional extras: pdf, ocr, transcription and jev. Start with the base installation; enable only the engines you need. Media, attachments, deployment. Native connection observations and testing limits are recorded in acceptance; discovery does not establish model workflow quality.

Develop and contribute

Report reproducible bugs and propose workflows in Issues. Read CONTRIBUTING; coding agents start with AGENTS.md. Run checks for the behavior you change. A substantial release gets one full run, one canonical build and artifact identity verification.

Teleloom v0.5.0 is the first public release, published with a fresh Git history. The Python package, module, CLI and MCP server are named teleloom; runtime skills use the teleloom- prefix. Install from this repository or its release artifacts; no PyPI distribution is claimed. Changelog · Documentation.

MIT licensed. Inspired by chigwell/telegram-mcp, implemented independently. Teleloom is an independent project, unaffiliated with Telegram.

Related MCP Connectors

Related MCP Servers

  • F
    license
    B
    quality
    D
    maintenance
    An MCP server that enables interaction with Telegram messaging platform, allowing users to retrieve unread messages, fetch entity information, and send messages through natural language interfaces.
    6
    10
    -
  • A
    license
    Not graded
    quality
    C
    maintenance
    An MCP server that enables interaction with Telegram to send, read, and search messages across chats and dialogs. It supports waiting for incoming messages and retrieving conversation history through natural language commands.
    15 npm
    3
    MIT
  • A
    license
    Not graded
    quality
    C
    maintenance
    Enables AI agents to interact with Telegram accounts through MCP, supporting messaging, contacts, groups, media, and admin functions.
    4
    Apache 2.0
  • F
    license
    B
    quality
    D
    maintenance
    Enables users to read, search, and manage Telegram messages in channels, groups, and private chats through MCP tools.
    8
    -