Agents & Pipeline Builder

​

Overview

Agents & Pipeline Builder is a built-in Module that connects a conversation agent to ELITEA’s own Model Context Protocol (MCP) server. Once enabled, the agent can create and update agents and pipelines directly from chat, while also discovering and calling other agents and toolkit operations that have been explicitly made available through MCP — all from within a single conversation, without the user needing to pre-wire every dependency upfront.

---

​

Prerequisites

• MCP Enabled: MCP features must be enabled on the platform. If the Agents & Pipeline Builder toggle does not appear in the Modules list, contact your Elitea administrator.
• Permission Level: User role with conversation or agent edit access
• Default Availability: The feature can be enabled by default for new conversations through the project-level Agent & Pipeline Builder setting.
• Agents tagged with mcp: To appear as callable tools, agents must have the mcp tag applied. Agents without this tag are not listed, regardless of publish status.
• MCP-Available Toolkits (optional): Toolkits with Available via MCP enabled in their settings, so their individual tools are callable.
• Elitea Personal Access Token (for internal MCP toolkits when required): Some internal MCP toolkit actions require a valid Elitea personal access token. If the token is missing or expired, renew or create one before retrying the action.

---

​

How It Works

When Agents & Pipeline Builder is enabled, the agent connects to ELITEA’s built-in MCP server, which exposes callable resources and supports chat-driven builder workflows: How a tool call works

1. The agent decides to call a resource — for example, PR_Review_Agent with task: "Review PR #42"
2. The MCP server resolves the name and routes the call:
   • Agent call → runs the full agent prediction pipeline and returns the agent’s final reply
   • Toolkit tool call → executes the tool with the provided arguments and returns the result as JSON
3. The response is returned to the calling agent, which incorporates it into its reasoning

---

​

Enabling Agents & Pipeline Builder

Agents & Pipeline Builder can be toggled on per conversation or saved as part of an agent’s configuration for persistent use across all conversations with that agent.

​

In a Conversation

Enable Agents & Pipeline Builder for an ad-hoc conversation.

1. Navigate to your conversation.
2. Locate the chat input toolbar at the bottom of the screen.
3. Click the + icon in the toolbar to open the menu.
4. Hover over Modules to reveal the flyout panel.
5. Find Agents & Pipeline Builder in the list and click the toggle to enable it.
6. A success toast notification appears: “Modules configuration updated”.
7. Click anywhere outside the panel to close it.

Once enabled, the agent has access to builder capabilities and all MCP-exposed tools in that conversation. The agent can create or update agents and pipelines, and call available MCP tools as needed based on the conversation flow and user instructions.

​

In Agent Configuration

Configure Agents & Pipeline Builder as part of an agent’s saved configuration so it is active for all new conversations using that agent.

1. Navigate to Agents in the main menu.
2. Select the agent you want to configure or create a new agent.
3. Within the TOOLS section, locate the MODULES subsection.
4. Find the Agents & Pipeline Builder toggle. If it is not visible, click Show all to expand the full list.
5. Click the Agents & Pipeline Builder toggle to enable it.
6. Click Save at the top of the configuration page.
7. New conversations created with this agent will have Agents & Pipeline Builder active by default.

---

​

Available Tools at Runtime

The exact tools the agent sees at runtime depend on your project’s configuration.

​

Agent Tools

Agents that have the mcp tag applied are exposed as callable tools. Each such agent produces one tool:

​

Toolkit Tools

Toolkits with Available via MCP turned on in their MCP Options section expose their selected tools individually:

---

​

Using Agents & Pipeline Builder

Once enabled, the agent automatically has access to all discovered tools. No additional prompting or configuration is needed — the agent uses them based on the task requirements.

​

Create an Agent from Chat

You can use Agents & Pipeline Builder to create a new agent directly from a conversation.

1. Open a conversation where Agents & Pipeline Builder is enabled.
2. Ask the agent to create a new agent and describe what it should do.
3. Include the key details in your request.
4. Review the generated result in chat.
5. If needed, ask the agent to refine or update the generated agent configuration.
6. Save or apply the created agent configuration in the platform when the result matches your intent.

​

Example Interaction

User: “List all open pull requests from the GitHub toolkit and ask the code review agent to review each one.” Setup required:

• GitHub toolkit has Available via MCP enabled and list_pull_requests in its selected tools
• The code review agent has the mcp tag applied

Behind the Scenes:

1. The agent calls My_GitHub_Toolkit_list_pull_requests (normalised from "My GitHub Toolkit" + list_pull_requests) to fetch open PRs
2. For each PR, the agent calls Code_Review_Agent with a self-contained task that includes the PR title and diff inline
3. The code review agent runs its full prediction pipeline and returns its findings as text
4. The calling agent collects all reviews and presents a consolidated summary to the user

---

​

Example Scenarios

Dynamic Agent Coordinator

Scenario: A meta-agent that delegates tasks to specialised agents at runtime, without pre-wiring every dependency in the configuration.Setup:

• Agents & Pipeline Builder: Enabled on the coordinator agent
• Reporting Agent and QA Agent both have the mcp tag applied

User Request:

Summarise the last sprint and send the report through the reporting agent, then run the regression tests using the QA agent.

Workflow:

1. Coordinator discovers available agent tools from the MCP server
2. Calls Reporting_Agent with task: "Generate a sprint summary for the last sprint. Include completed items, blockers, and carry-overs."
3. Receives the sprint report as text
4. Calls QA_Agent with task: "Run the regression test suite and return the pass/fail summary."
5. Receives the test results
6. Combines both outputs and returns a single status update to the user

The coordinator never needed hardcoded references to the child agents — adding a new agent with the mcp tag makes it immediately available.

Release Automation Agent

Scenario: A DevOps engineer wants an agent that bridges Jira and GitHub using natural language — finding tickets, merging their PRs, and updating statuses.Setup:

• Agents & Pipeline Builder: Enabled
• Dev GitHub toolkit — Available via MCP on; selected tools include list_pull_requests, merge_pull_request
• Project Jira toolkit — Available via MCP on; selected tools include search_issues, update_issue

Exposed tool names (after normalisation):

• Dev_GitHub_list_pull_requests, Dev_GitHub_merge_pull_request
• Project_Jira_search_issues, Project_Jira_update_issue

User Request:

Find all Jira tickets marked "Ready for Release" and merge the associated GitHub pull requests.

Workflow:

1. Agent calls Project_Jira_search_issues with JQL status = "Ready for Release"
2. For each ticket, extracts the linked PR number from the ticket fields
3. Calls Dev_GitHub_merge_pull_request for each PR
4. Calls Project_Jira_update_issue to transition each ticket to “Released”
5. Returns a summary: merged PRs and updated Jira tickets

QA Coordinator Agent

Scenario: A QA team uses one coordinator agent to orchestrate test execution and bug filing across dedicated specialist agents and a Jira toolkit — all resolved at runtime.Setup:

• Agents & Pipeline Builder: Enabled on the coordinator agent
• Test Runner Agent — mcp tag applied; runs the test suite and returns results
• Failure Analysis Agent — mcp tag applied; classifies failures as known regressions or new bugs
• QA Jira toolkit — Available via MCP on; selected tools include create_issue

User Request:

Run today's regression suite, analyse any failures, and log new bugs.

Workflow:

1. Coordinator calls Test_Runner_Agent with task: "Run the full regression suite and return pass/fail per test case."
2. Agent receives structured results with failure details
3. Coordinator calls Failure_Analysis_Agent with task: "Classify the following failures as known regressions or new bugs: {failures_list}"
4. Analysis agent returns a categorised list
5. For each new bug, coordinator calls QA_Jira_create_issue with the failure details
6. Coordinator sends a final summary: tests run, failures found, bugs filed

---

​

Troubleshooting

Agents & Pipeline Builder toggle is not visible in the Modules list

The toggle only appears when MCP features are enabled at the platform level. Contact your Elitea administrator and ask them to enable MCP exposure and MCP in menu in the platform configuration.

No agent tools appear after enabling

Agents are only listed if they have the mcp tag applied. Publish status alone is not sufficient.To expose an agent:

1. Open the agent in the Agents menu
2. Go to the agent’s settings and add the mcp tag
3. Save the agent

The agent will now appear as a tool when Agents & Pipeline Builder is enabled.

No toolkit tools appear after enabling

Toolkit tools are only exposed when the toolkit has Available via MCP enabled in its configuration. Go to Toolkits, open the toolkit, locate the MCP Options section, and enable the toggle. Only selected_tools within that toolkit will be exposed — verify the correct tools are selected.

The agent calls the wrong tool or gets confused by similar tool names

Tool names are derived from display names: any character outside [A-Za-z0-9_-] is replaced with _. If two resources produce the same normalised name, only the first is exposed and the second is silently skipped.Make names clearly distinct. For example, "GitHub - Frontend Repo" and "GitHub - Backend Repo" normalise to different names, while "GitHub (Frontend)" and "GitHub Frontend" both become GitHub_Frontend_ and may collide.Rename conflicting agents or toolkits and verify the resulting normalised names are unique before relying on them.

Agent tool calls time out or return errors

Agent tool calls run the full agent prediction pipeline, which can take significant time for complex agents. If timeouts occur:

• Ensure the target agent’s LLM is healthy and its configuration is complete
• Check if the target agent has its own toolkits that may be failing to initialise
• Simplify the task argument — the called agent must be able to complete it independently

Toolkit tool calls return 'Access denied' or similar errors

Toolkit tool execution uses the current user’s session credentials. Verify:

• The user has access to the project containing the toolkit
• The toolkit’s stored credentials (API keys, tokens) are valid and not expired
• The toolkit’s Available via MCP option is still enabled

Internal MCP toolkit actions ask for a personal access token

Some internal MCP toolkit actions require a valid Elitea personal access token.If the token is missing or expired:

• Create a new Elitea personal access token
• Renew the existing token if it has expired
• Retry the action after updating the token

If the issue persists, verify that the token belongs to the current user and has access to the current project.

---

​

Related Features