Bel Esprit MCP Tools

Bel Esprit is an aiXplain MCP server that exposes a set of tools for discovering aiXplain assets, onboarding custom code, and building, deploying, and running agents — all from an MCP client. Use it to search the marketplace, turn a Python function into a reusable tool, and generate a working agent from a schema without leaving your MCP-enabled environment.

Authentication

All tools require an aiXplain API key, supplied one of two ways:

• The x-api-key HTTP header on each request, or
• The TEAM_API_KEY or AIXPLAIN_API_KEY environment variable.

Get an API key ↗

---

Search Integrations

search_integrations searches for integrations and models on aiXplain (e.g. Gmail, Slack, Google Drive, databases, third-party APIs requiring OAuth or API key auth).

Parameter	Type	Default	Description
query	string	required	Descriptive search query
top_k	int	50	Max results to return

Returns: List of integrations including id, name, description, auth_scheme, requires_credentials, credential_inputs, and path.

---

Search Tools

search_tools searches the aiXplain marketplace for ready-made capability tools (e.g. Tavily Web Search, Firecrawl, Code Execution, Docling).

Parameter	Type	Default	Description
query	string	required	Descriptive search query
top_k	int	20	Max results to return

Returns: List of tools including id, name, description, and path.

---

Search Scripts

search_scripts searches for custom user-written scripts already deployed on the aiXplain platform.

Parameter	Type	Default	Description
query	string	required	Search query
page_size	int	20	Max results to return

Returns: List of user scripts including id, name, description.

---

Onboard Code

onboard_code deploys a Python function as a reusable utility tool on aiXplain. Takes 10–60 seconds.

Parameter	Type	Default	Description
code	string	required	Python source code (all parameters must have type hints)
name	string	required	Tool name — max 24 characters
description	string	required	What the tool does

Returns: Deployed tool metadata including id.

---

Generate Agent

generate_agent builds, deploys, and optionally smoke-tests an aiXplain agent from a schema.

Parameter	Type	Default	Description
name	string	required	Agent name
description	string	required	What the agent does
instructions	string	required	System prompt — role, capabilities, tone, constraints
connections	list	null	From search_integrations results. Each item: {id, name, description, selection_rationale, auth_scheme, requires_credentials, credential_inputs, credentials}. When requires_credentials is true, the user must supply actual credential values in the credentials dict before calling this tool.
tools	list	null	From search_tools or search_scripts results
script_suggestions	list	null	Custom Python tools to create inline
llm_id	string	GPT-4o	LLM model ID
deploy	bool	true	Deploy after creation
verify	bool	true	Smoke-test after deploy
verification_query	string	"Hello, confirm you are operational."	Query used for smoke test
require_all_tools	bool	false	Fail if any tool is missing
create_scripts	bool	true	Build script_suggestions before deploying

Returns: Deployed agent result including agent_id.

---

Run Agent

run_agent runs a deployed aiXplain agent and returns its response.

Parameter	Type	Default	Description
agent_id	string	required	Agent ID from generate_agent or any deployed agent ID
query	string	required	Input query to send to the agent

Returns: Agent response text.

---

Typical workflow

The tools are designed to chain into an end-to-end "build me an agent" flow:

1. search_integrations / search_tools / search_scripts — discover the assets the agent should use, and read each result's auth_scheme / requires_credentials.
2. onboard_code — optionally turn a custom Python function into a reusable tool.
3. generate_agent — assemble the selected connections, tools, and script_suggestions into an agent, then deploy and smoke-test it. Supply credential values for any connection whose requires_credentials is true.
4. run_agent — send queries to the deployed agent using the returned agent_id.