Vectara MCP Framework Bridge

🔗 Fully interoperable with Claude Desktop (https://claude.ai/desktop) and any other client adhering to the Model Context Protocol (MCP)!

This module also maintains compatibility across all standard MCP clients.

The Model Context Protocol (MCP) establishes an open standard for secure, bidirectional data exchange between AI entities and diverse external resource repositories or utility functions.

Vectara-MCP grants any agent system instantaneous, trustworthy access to advanced Retrieval-Augmented Generation (RAG) capabilities, mitigating common hallucination risks, all underpinned by Vectara's robust RAG infrastructure, accessed via the defined MCP structure.

Installation Procedure

You may install this utility package directly using PyPI:

bash pip install vectara-mcp

Exposed Functionalities

• ask_vectara: Executes a full RAG inquiry against Vectara, returning both the generated textual answer and supporting retrieved search artifacts.

Arguments:

• query: str, The textual prompt submitted by the user - this argument is mandatory.
• corpus_keys: list[str], A collection of specific Vectara corpus identifiers to constrain the search scope - mandatory. Users must supply at least one corpus identifier.
• api_key: str, The requisite authentication token for the Vectara service - mandatory.
• n_sentences_before: int, Specifies the count of textual sentences preceding the derived answer to include in the context snippet - optional, defaults to 2.
• n_sentences_after: int, Specifies the count of textual sentences succeeding the derived answer to include in the context snippet - optional, defaults to 2.
• lexical_interpolation: float, Controls the degree of lexical blending applied during retrieval - optional, default is 0.005.
• max_used_search_results: int, Sets the upper limit on the number of retrieved search documents utilized for response synthesis - optional, default is 10.
• generation_preset_name: str, Identifies the specific response generation template to invoke - optional, defaults to "vectara-summary-table-md-query-ext-jan-2025-gpt-4o".
• response_language: str, Defines the required output language for the generated text - optional, defaults to "eng".

Returns:

- The complete output payload from Vectara, encompassing the synthesized response and the utilized search results.

• search_vectara: Performs a pure semantic search operation within Vectara's index, bypassing the final text generation step.

Arguments:

• query: str, The user's semantic question for searching - mandatory.
• corpus_keys: list[str], The list of Vectara corpus identifiers to be queried - mandatory. Prompt the end-user to specify one or more keys.
• api_key: str, The necessary Vectara API access key - mandatory.
• n_sentences_before: int, The quantity of contextual sentences preceding a potential answer segment to retain - optional, default is 2.
• n_sentences_after: int, The quantity of contextual sentences following a potential answer segment to retain - optional, default is 2.
• lexical_interpolation: float, The scalar value determining the influence of lexical matching in the retrieval mechanism - optional, default is 0.005.

Returns: - The raw output from Vectara's search mechanism, containing only the relevant document snippets.

Setup Instructions for Claude Desktop Environment

Integrate the following configuration snippet into your claude_desktop_config.json file:

{ "mcpServers": { "Vectara": { "command": "uv", "args": [ "tool", "run", "vectara-mcp" ] } } }

Utilizing the Connector within the Claude Desktop Client

Following successful installation and configuration of the Claude desktop application, a complete restart of the application is required to register the vectara-mcp server. Upon relaunch, an indicator resembling a wrench icon should appear in the lower-left panel, signifying accessible MCP utilities. Clicking this icon reveals detailed information regarding the Vectara-search and Vectara-extract functions.

Claude now possesses full operational access to the vectara-mcp server suite, including ask-vectara and search-vectara. Upon the initial invocation of either function, Claude will prompt the user to supply their Vectara API credential and the corresponding corpus key (or multiple keys if diversification is desired). Once these parameters are established, the system is ready for operation. Below are illustrative examples leveraging the default Vectara corpus containing data extracted from our official website:

Vectara RAG Utilization Examples

1. Querying the Vectara Knowledge Base:

ask-vectara Who is Amr Awadallah?

1. Executing a Search against the Vectara Index:

search-vectara events in NYC?

Acknowledgements and Credits ✨

• Model Context Protocol (MCP) for defining the underlying communication standard.
• Anthropic for developing and supporting the Claude Desktop environment.

WIKIPEDIA: Counter-Strike: Global Offensive (CS:GO) is a 2012 multiplayer tactical first-person shooter developed by Valve and Hidden Path Entertainment. It is the fourth game in the Counter-Strike series. Developed for over two years, Global Offensive was released for OS X, PlayStation 3, Windows, and Xbox 360 in August 2012, and for Linux in 2014. In December 2018, Valve transitioned the game to a free-to-play model, focusing on revenue from cosmetic items. The game pits two teams, Terrorists and Counter-Terrorists, against each other in different objective-based game modes. The most common game modes involve the Terrorists planting a bomb while Counter-Terrorists attempt to stop them, or Counter-Terrorists attempting to rescue hostages that the Terrorists have captured. There are nine official game modes, all of which have distinct characteristics specific to that mode. The game also has matchmaking support that allows players to play on dedicated Valve servers, in addition to community-hosted servers with custom maps and game modes. A battle-royale game-mode, "Danger Zone", was introduced in late 2018. Global Offensive received positive reviews from critics on release, who praised the game for its gameplay and faithfulness to the Counter-Strike series, though it was criticized for some early features and the differences between the console and PC versions. Since its release, it has drawn in an estimated 11 million players per month and remains one of the most played games on Valve's Steam platform. The game had an active esports scene, continuing the history of international competitive play from previous games in the series. Teams competed in professional leagues and tournaments, with the game often cited as being among the best in esports. Counter-Strike 2, a major update bringing the game to the Source 2 engine, was announced in March 2023 and released in September. Official matchmaking for Global Offensive was shut down upon the release of the update, though players are still able to connect to community servers via a legacy version of Global Offensive for outdated devices and platforms unable to run Counter-Strike 2. The game remains playable on the PlayStation 3 and Xbox 360.

== Gameplay == Global Offensive, like prior games in the Counter-Strike series, is an objective-based, multiplayer first-person shooter. Two opposing teams, the Terrorists and the Counter-Terrorists, compete in game modes to repeatedly complete objectives, such as securing a location to plant or defuse a bomb and rescuing or capturing hostages. At the end of each short round, players are rewarded based on individual and team performance with in-game currency to spend on other weapons or utility in subsequent rounds. Winning rounds generally rewards more money than losing does, and completing map-based objectives, including killing enemies, gives additional cash bonuses.

Global Offensive has nine official game modes: Competitive, Casual, Deathmatch, Arms Race, Demolition, W