Every couple of weeks, Mark walks through the best questions, workarounds, and discoveries from the Tana community Slack. Week 8 leans into MCP and fields: telling an agent where a node lives, hiding fields without losing them, untangling duplicate fields, a deeper subtree filter, fixing transcript speaker names, and automatically stamping who created a task.
Point an MCP agent at a node [0:01]
Harris asked how to tell a coding agent where a node is. The simplest answer: copy node link (from command + K or the right-click menu) and paste it into the agent. With the local API MCP connected, the agent reads exactly that node. It recognizes attached files by type, though it can't yet view or OCR images.
Truly hidden metadata fields [1:44]
Alberto found the field-is-metadata option on the set field values command. It pushes a field into the debug menu, completely out of sight, while you can still filter by it in search. It's a more thorough alternative to the usual hidden fields (collapsed at the top of a node) for power users who want zero visual clutter. The trade-off: if a search misbehaves, the field isn't visible on the node, so the cause isn't obvious.
Stop duplicating the same field [5:46]
A recurring confusion (raised by Martin and Sabina): creating a field with the same name again makes a brand-new field rather than reusing the existing one, which breaks searches and inheritance. To reuse a field, pick it from the existing options on the supertag rather than retyping it. To check whether two fields are actually the same, paste references to both into one node: if Tana refuses the second as a duplicate they're the same; if it accepts both they're different. Highlight the two and run command + K then merge node to combine them, and inheritance starts working.
An extended child-of filter [9:06]
Michael wanted to exclude a node's whole subtree to any depth, which the child-of field doesn't really do. Use parents descendants with refs instead, negated, to filter out everything in the tree under a given node. There's a nuance: a search node counts its own results as descendants on the next run, so add parents descendants with non-search refs (an older, lightly documented operator) to ignore the search node's own values when filtering.
Rename transcript speakers [13:11]
Also from Michael: you can't edit speaker names directly in a transcript. Open the transcript node, run command + K then debug, and scroll to the bottom to find the speaker names. Rename them there and the corrected names flow into the summary, and into meeting notes if you reference them.
Auto-assign a task to its creator [14:48]
Harald wanted tasks assigned to whoever created them, wherever they're made (something the Tana team uses internally). Tana tracks creation in a hidden edited-by system value, but you can't search those user values directly. The trick is an AI field that reads edited-by and returns the matching person from a team search node. Add a run AI fields command to the tag's on-added automation and the creator fills in automatically the moment a task is tagged.