indico-mcp

indico-mcp

An MCP (Model Context Protocol) server for the Indico meeting agenda system, letting AI agents search event categories, browse agendas, and extract contribution and session details.

Works with any Indico instance — configure multiple instances simultaneously and switch between them per tool call.

Tools

All tools accept an optional instance parameter to select which Indico server to query.

Requirements

• Python 3.11+
• uv

Installation

git clone <repo>
cd indico-mcp
uv sync

Configuration

Copy the example env file and fill in your details:

cp .env.example .env

Single instance

INDICO_BASE_URL=https://indico.fysik.su.se
INDICO_TOKEN=indp_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

Multiple instances

INDICO_INSTANCES=cern,su
INDICO_DEFAULT=su

INDICO_CERN_URL=https://indico.cern.ch
INDICO_CERN_TOKEN=indp_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

INDICO_SU_URL=https://indico.fysik.su.se
INDICO_SU_TOKEN=indp_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

INDICO_DEFAULT sets which instance is used when no instance= argument is passed to a tool. Tokens are optional for public-only access.

Room booking setup

Room booking tools (search_rooms, find_available_rooms, book_room) require knowing the site name as configured in Indico's room booking module. These names are not discoverable via the API — you must look them up in the Indico room booking interface (they appear as location headings, e.g. "AlbaNova Main Building", "Albano Building 3").

1. Configure site names

Add the known site names to your .env:

INDICO_SU_ROOM_LOCATIONS=AlbaNova Main Building,Albano Building 3

For single-instance: INDICO_ROOM_LOCATIONS=Main Building,Building 40

2. Run room discovery

Once site names are configured, run discover_rooms once (e.g. by asking your agent to do it). This scans reservation history across all configured sites and saves a local room catalogue to ~/.indico_mcp/{instance}_rooms.json (override the directory with INDICO_ROOMS_CACHE_DIR).

After discovery, search_rooms finds rooms by partial name across all catalogued sites without needing a location argument. Rooms with no booking history are found on-the-fly via Indico's room name search endpoint, so discovery does not need to be exhaustive.

Re-run discover_rooms periodically (or whenever new rooms are added) to keep the catalogue fresh. The agent will offer to run it automatically if the cache file is missing when a room operation is requested.

Getting a token

1. Log in to your Indico instance
2. Go to My Profile → Personal Tokens (or /user/tokens/)
3. Click Create token
4. Give it a name and enable the Classic API (read) scope — this grants read access to the event export API
5. If you also need write access (creating events, bookings), enable Classic API (read and write) instead
6. Copy the generated token (starts with indp_)

Note on scope naming: The scope is labelled "Classic API" in the UI but this refers to Indico's HTTP Export API (/export/), which is the only way to programmatically read event and contribution data. Despite the name, using it with a modern Bearer token (indp_...) is the current recommended approach — the deprecated system is the old API key (a separate, shorter token found under My Profile → HTTP API).

Connecting to Claude

Run claude add indico -- uv run --directory /path/to/indico-mcp indico-mcp or manually add to your Claude config (~/.claude.json or Claude Desktop settings):

{
  "mcpServers": {
    "indico": {
      "command": "uv",
      "args": ["run", "--directory", "/path/to/indico-mcp", "indico-mcp"]
    }
  }
}

The .env file in the project directory is loaded automatically. Alternatively, pass configuration directly via the env block in the MCP config:

{
  "mcpServers": {
    "indico": {
      "command": "uv",
      "args": ["run", "--directory", "/path/to/indico-mcp", "indico-mcp"],
      "env": {
        "INDICO_INSTANCES": "cern,su",
        "INDICO_DEFAULT": "su",
        "INDICO_CERN_URL": "https://indico.cern.ch",
        "INDICO_CERN_TOKEN": "indp_xxx",
        "INDICO_SU_URL": "https://indico.fysik.su.se",
        "INDICO_SU_TOKEN": "indp_yyy"
      }
    }
  }
}

Usage examples

# Events in the next month (default instance)
search_category_events(from_date="2025-04-01", to_date="2025-04-30")

# Events in a specific category on CERN Indico
search_category_events(category_id=72, instance="cern")

# First page of contributions for a meeting
get_event_contributions(event_id=1234567, instance="cern")

# Continue with paging using `pagination.next_offset`
get_event_contributions(event_id=1234567, limit=100, offset=0, instance="cern")

# Full session/agenda structure of a conference
get_event_sessions(event_id=9876543, instance="su")

# Full-text search
search_events_by_keyword("dark matter", instance="cern")

# Navigate the category hierarchy
list_category_info(category_id=0, instance="su")

# First page of attachments for an event
list_event_attachments(event_id=1234567, instance="cern")

# Continue with paging using `pagination.next_offset`
list_event_attachments(event_id=1234567, limit=100, offset=0, instance="cern")

# List attachments for a specific contribution
list_event_attachments(event_id=1234567, contribution_id=42, instance="cern")

# Download a file (URL from list_event_attachments output)
download_attachment(download_url="https://indico.cern.ch/event/.../file.pdf")

How it works

The server uses the Indico HTTP Export API (/export/) with detail=contributions and detail=sessions query parameters to retrieve structured agenda data. Authentication uses a standard Authorization: Bearer <token> header. The /api/ endpoints in Indico are write-only (POST); all read operations go through /export/.

File attachments are discovered via the folders structure included in the export API response, which provides direct download URLs, filenames, content types, and sizes. Downloads are authenticated with the same Bearer token and saved locally (to a temp directory by default, or a specified path). Maximum file size is 100 MB.

Data Sensitivity & Privacy

Indico agendas and their attachments may contain sensitive or confidentialinformation — internal meeting minutes, unpublished research results,restricted-access talks, personnel matters, and similar content.

What gets sent to the LLM provider

When an AI assistant calls tools such as get_event_details,get_event_contributions, get_event_sessions, or list_event_attachments,the returned data (titles, abstracts, speaker names, descriptions) is forwardedto the underlying language model. If you are using a cloud-hosted LLM provider(e.g. Anthropic Claude, OpenAI GPT), this data leaves your local machine andis processed by the provider according to their data-handling policies.

Similarly, when download_attachment is used and the AI reads the downloadedfile, the file content is forwarded to the LLM provider.

Before processing agenda data with an AI assistant:

• Check whether the event or category is internal or access-restricted.
• Verify your organisation's data-handling and AI-usage policies.
• For confidential or proprietary material, prefer a locally-hosted LLM.
• Attachments with is_protected: true in list_event_attachments output areexplicitly access-restricted by the owner — require explicit user approvalbefore downloading or summarising them.

Token and access control

The Indico API token grants read (and optionally write) access to the instance.Apply the principle of least privilege:

• Use a read-only token (Classic API read scope) unless write access isexplicitly needed.
• Do not store tokens in version-controlled files (use .env, which is excludedby .gitignore).
• Rotate tokens if they are accidentally exposed.

Room booking

The book_room tool creates real reservations immediately. Always confirmall booking details with the user before calling book_room with dry_run=False,and never retry automatically on network failures — each retry may create aduplicate booking.

Contributing

Feature requests and bug reports are welcome. Contributions are especially encouraged from Indico users who can test new functionality against a real instance before submitting a pull request — the Indico API has enough instance-to-instance variation that untested changes are hard to review reliably.

Substantial contributions will be recognised by adding the contributor as an author.

License

Copyright (c) 2026 Christian Ohm. MIT License — see the LICENSE file.

Indico itself is also MIT licensed, so this can perhaps be absorbed into that repository.