Agent
Capabilities

MCP Servers

MCP servers give your agents access to external tools and data sources — web search, file systems, databases, APIs, and more. Taskeract lets you browse, install, and manage them from Settings > MCP Servers or from a project’s MCP tab.

Installing

Browse servers by category or search by name. Click a server to read its README and see auto-detected install options. Choose an install option to open the add form with the command and arguments pre-filled, or click Add Manually to fill in the details yourself. The + button next to the search field also opens the manual add form. If the server needs API keys or other credentials, add them as environment variables — credentials are encrypted and stored locally on your machine.

Servers installed from Settings are available to all agents across every project. Servers installed from a project’s MCP tab are scoped to that project and shared with teammates — anyone who opens the project will see them listed as recommendations and can install them with one click.

Environment Variables

If your project uses .env files (managed from the Sync Files tab), their variables are automatically available to your MCP servers. When configuring a server, use $${VAR_NAME} to reference a variable — for example, $${OPENAI_API_KEY}. You can mix references with other text, like http://localhost:$${PORT}/api. For web app projects, PORT and SESSION_URL are also available. When your variables change, servers are updated automatically.

Auto-Approve Tools

By default, agents ask for permission before running MCP tools. If you trust an agent to run tools on its own, enable Auto-approve tools in Settings > Agents for that agent. When enabled, the agent runs tools without prompting you each time. This setting is per agent, so you can enable it for some agents while keeping manual approval for others.

Managing

Your installed servers appear in the left panel. Click one to view its details, update credentials with Configure, or remove it with Uninstall. The MCP Servers tab also works when connected via web remote.

Wiki Integration

Agents can read and search your project's wiki directly. This gives agents access to team knowledge — onboarding guides, architecture decisions, runbooks — without needing to leave their workflow. Content is returned as markdown text, and agents can share browser URLs so you can click through to the original page.

Supported Wikis

• GitHub, GitLab, Bitbucket — the wiki tied to your project's git remote. Detected automatically from the repository URL. GitHub and Bitbucket wikis are cloned locally for fast access; GitLab uses its REST API.
• Confluence — Atlassian Confluence, using the same credentials as your Jira issue tracker. Requires a confluence_space_key in your project config.
• Taiga — Taiga wiki pages, using your Taiga issue tracker credentials.

How It Works

Your agent can interact with the wiki in two ways:

• Read — retrieves a specific page by its slug or title. Returns the full page content as markdown along with a URL you can open in your browser.
• Search — searches across all wiki pages. Returns matching page titles, snippets, and URLs.

For GitHub and Bitbucket wikis, the first access clones the wiki repository locally. Subsequent reads are instant. The wiki is kept in sync automatically in the background.

Setup

No setup is needed for GitHub, GitLab, or Bitbucket wikis — they work automatically if you have a token saved for the hosting provider in Settings > Accounts.

For Confluence, add confluence_space_key to your project's .tact/project.yaml under the issue_tracker section (see Configuration). Your Jira account credentials are reused for Confluence access.

For Taiga, wiki access works automatically when a Taiga issue tracker is configured for the project.

Library Source Resolver

AI agents often only see compiled or bundled dependency artifacts — transpiled JavaScript in node_modules, .class files, or .dlls. The Library Source Resolver fetches the original source code for any library on demand so agents can read the actual implementation instead of guessing.

How It Works

When an agent needs to understand a library’s internals, it asks Taskeract for the library name and version. Taskeract resolves the source repository from the package registry, downloads the source at the matching version, and returns a filesystem path. The agent then explores the source directly — searching for functions, reading implementations, and understanding the library’s real API.

If the library was already fetched, the cached path is returned instantly.

Supported Ecosystems

• npm — resolves from the npm registry (e.g. express, lodash, @angular/core).
• PyPI — resolves from the Python Package Index (e.g. django, requests, flask).
• Maven — resolves from Maven Central (e.g. jackson-databind, guava).
• NuGet — resolves from the NuGet gallery (e.g. Newtonsoft.Json, Dapper).
• RubyGems — resolves from rubygems.org (e.g. rails, sidekiq).

Taskeract detects the project’s ecosystem automatically and tries the matching registry first. If the library isn’t found there, all other registries are tried as a fallback.

Caching

Cloned sources are cached on disk so repeated requests for the same library and version are instant. The cache is shared across all projects — if two projects both use express@4.18.2, the source is only cloned once. Older entries are cleaned up automatically to manage disk space.

Excavator

The Excavator launches one or more AI agents to systematically scan your project for real bugs. Agents run continuously in the background, reporting confirmed issues with file locations, line numbers, and severity levels. They coordinate so multiple agents don’t duplicate findings.

Starting a Scan

Click the Excavator icon in the session header to open the sidebar. Each configured agent is listed with a count stepper (0–3), so you can run multiple agent types simultaneously — for example, one Claude Code and two Codex agents. Click Start to begin scanning.

Agents run in non-interactive mode and loop automatically. After each round, an agent checks what’s already been found and continues scanning for new bugs until you stop them.

Reviewing Results

Bugs appear in the sidebar as they’re found, with a severity indicator (critical, high, medium, low, or info). Click a bug to see the full description rendered as markdown, along with the file path and line range.

Existing bug-type issues from your issue tracker are fetched at the start so agents don’t report duplicates.

Taking Action

For each bug, you can:

• Promote to Tracker — automatically creates an issue in your connected issue tracker (GitHub, GitLab, Jira, Linear, Trello, or Taiga) with the correct bug type and label.
• Copy as Prompt — copies the bug details to your clipboard, ready to paste into an agent session for fixing.
• Dismiss — hides the bug from the list if it’s not relevant.

Use checkboxes to select multiple bugs for batch promote, copy, or dismiss. Promoted bugs are marked with a checkmark icon and excluded from future batch promotes, so you can safely leave them checked while working through the list. The sidebar supports full keyboard navigation — see Keyboard Shortcuts for details.

Stopping

Click Stop to deactivate the Excavator. Agents finish their current round before stopping. If you need to halt them immediately, click Force Stop to kill the agents without waiting. The bug list is preserved until you close the sidebar or clear it manually.