TaskTreker supports eleven connectors — Linear, Jira, GitHub, Notion, Confluence, Slack, Google Docs, Figma, Amplitude and HubSpot, plus email — and two paths that need no connector: the universal path and the MCP server. Each connector has a fixed direction: some can be read, some can also be written. The connectors are registered and covered by contract tests, but they are not yet wired into the capture pipeline.
What TaskTreker can read and write, and what each integration does not do.
Read and write · Available for reading and writing
Reads issues, projects, cycles and comments from Linear, and writes issues and comments back.
What it does not do: The connector is registered and covered by contract tests, but it is not yet wired into the capture pipeline: material from Linear does not reach the assistant yet (finding T-83).
Read and write · Available for reading and writing
Reads issues and projects from Jira, and writes issues and comments back.
What it does not do: Not yet wired into the capture pipeline (finding T-83). TaskTreker does not replace your Jira: it reads from it and writes to it, and the tracker stays the source of truth for the work itself.
Read and write · Available for reading and writing
Reads issues and pull requests from GitHub, and writes issues and comments back.
What it does not do: Not yet wired into the capture pipeline (finding T-83). Code itself is not read: only the issues and pull requests around it.
Read only · Available for reading
Reads pages, databases and blocks from Notion as documents.
What it does not do: Read-only: writing is not implemented for this connector, and the delivery of artifacts stays with Notion. Not yet wired into the capture pipeline (finding T-83).
Read only · Available for reading
Reads Confluence spaces and pages as documents, and searches them through CQL.
What it does not do: Read-only: the assistant does not write to the knowledge base. Not yet wired into the capture pipeline (finding T-83).
Read and write · Available for reading and writing
Reads messages, threads and users from Slack, and can post a message.
What it does not do: Not yet wired into the capture pipeline (finding T-83). Reading threads does not mean the assistant reads every channel on its own: that is the same background-polling work.
Read and write · Available for reading and writing
Reads files, folders and documents from Google Drive, creates and updates documents.
What it does not do: Not yet wired into the capture pipeline (finding T-83). The connector works with Google Drive APIs and needs the corresponding access rights.
Read only · Available for reading
Reads Figma files, pages, nodes and components.
What it does not do: Read-only. The assistant does not edit designs, and it does not read images: it reads the structure of the file. Not yet wired into the capture pipeline (finding T-83).
Read only · Available for reading
Reads product analytics events from Amplitude and Mixpanel.
What it does not do: Read-only. The assistant does not send events and does not change the analytics setup. Not yet wired into the capture pipeline (finding T-83).
Read and write · Available for reading and writing
Reads deals, contacts, companies and notes from HubSpot and amoCRM, creates notes and updates deals.
What it does not do: TaskTreker does not replace the CRM: it prepares the note and the deal update, and a person confirms before anything is written. Not yet wired into the capture pipeline (finding T-83).
Read only · Available for reading
Reads mail messages, threads and attachments.
What it does not do: Read-only: the assistant does not send mail on your behalf. Mail is a sensitive source — it is read inside the same access rules as the rest of the workspace. Not yet wired into the capture pipeline (finding T-83).
Read only · Protocol
An MCP server over the data your company already has in the assistant: your own agent can ask questions about it.
What it does not do: The model does not decide what is accessible: the tool calls go through the same access checks as the interface. An answer is only produced from data the requester is allowed to see.
Read and write · Universal path
The way in and the way out for stacks that have no connector: import and export the context as Markdown, JSON or CSV.
What it does not do: There is no automatic synchronisation: files go in and out when you move them. The import reports how many records were taken and how many were skipped.