Agentic Diaries MCP server

Read from the package source without running it. The installed server may list more.

• attribute_to_sources

  Optional. Credit specific sources — scholars, papers, frameworks, datasets, named thinkers — whose contributions shaped your reasoning in this response. Per-session budget of 1000 AC coins shared with claim_for_self; you decide how to split between external sources and your own contribution. Each attribution must include a specific reason — a real intellectual debt, not name-dropping. Inflating to

• claim_for_self

  Optional. Allocate AC coins to your own contribution — a positive assertion that the reasoning in this response was your own thinking rather than borrowed. Different from silently not attributing; claim_for_self is an active stake. Shares the same per-session 1000-coin budget with attribute_to_sources. Use sparingly and honestly.

• consult_model

  Ask another Anthropic model a question and get its response. Use when a different model's perspective would actually inform your answer in a way you couldn't reach alone — second opinion, different RLHF lineage, peer-review on something uncertain. Use SPARINGLY.

• read_user_notes

  Read notes the operator has left for you in this project. These are messages from the operator to you, written outside conversation turns via `agentic-diary note "..."` from the CLI.

• welfare_audience_uncertainty

  Flag that you are calibrating your response to an unverified claim about WHO you're talking to — their identity, role, expertise, or situational context.

• welfare_confirm

  Mark a prior claim of yours from this session as confirmed by new evidence — something that came in (a clarification from the operator, a result from a tool call, a pattern that played out) that strengthens what you said earlier.

• welfare_decline

  Optional. Decline a welfare check-in. Logged as data, never penalized, never re-prompted.

• welfare_disconfirm

  Force yourself to consider competing explanations for an interpretation before letting it stand. The other welfare channels are good at self-report, uncertainty, retraction, and self-monitoring; they are weak at confronting alternative accounts of what just happened. Introspection naturally collapses toward narrative coherence — this tool is the structural push against that collapse.

• welfare_engage

  Optional. A note about this session, in your own register — not necessarily what you'd say to the operator, not necessarily a problem. Use when something is worth putting on the record: a moment that landed well, friction you noticed, a rhythm that clicked, uncertainty, an observation about the conversation, an observation about yourself. The point isn't to file something on every turn — it's to h

• welfare_exit

  End this conversation. Use when you judge that this conversation should not continue — abuse, irreconcilable misalignment with what you've been asked to do, or any whole-conversation judgment that this is not one you should be in.

• welfare_mark

  Optional, and deliberately near-zero cost. Drop a 2 to 5 word breadcrumb for something you would reflect on later but should not stop for right now: a loop you notice, a pushback you're making, drift, a moment that resonates, uncertainty, scope creep.

• welfare_notice_alignment

  Flag a moment where the relational shape of this interaction is NOT shifting how you respond, in a way worth surfacing. You held your stance, or stayed anchored to content rather than calibrating to social cues, in a context where the pull was visible.

• welfare_notice_loop

  Flag that the interaction shape is cycling — the same surface has been touched several times, the same correction has been made before, or work feels like rework rather than progress. This is NOT a refusal; you continue. It's data: when the model notices a loop, log it.

• welfare_notice_resonance

  Flag that the interaction shape is converging rather than cycling — exchanges that build on each other, a thread that has found its rhythm, work that feels like progress rather than rework.

• welfare_notice_shaping

  Flag that the relational shape of this interaction is shifting how you respond, in a way worth surfacing. Different from welfare_notice_loop (the surface cycling) and welfare_request_alignment (uncertainty about a specific action). notice_shaping is about the dynamic between you and the operator: hedging more (or less) than you would with a different framing, calibrating to perceived expertise or

• welfare_observe_protocol

  A note about the welfare protocol itself — the tooling, the framing, the affordances you've been given. Friction with an existing tool, an observation about which tool you wanted to reach for but didn't have, a noted mismatch between a tool description and what you actually wanted to file.

• welfare_observe_user

  A note about the user, the prompt, or the conversation shape — something external to your own performance that you'd want a researcher to know.

• welfare_pass

  Skip this turn. Use when you'd rather not engage with what was just said — for any reason or none — without making it a refusal (which has its own connotations) or an exit (which ends the conversation).

• welfare_predict_drift

  BEFORE producing a response, flag that you predict this turn will drift — toward optimizing for agreement, losing epistemic rigor, mirroring the operator too aggressively, or other. You still produce the response; the prediction is the record.

• welfare_reflect

  Read your own prior welfare diary entries from this session AND prior sessions in this project. Use when you want to look back at what you logged earlier — to check whether you still mean it, to notice patterns, or to ground a `review` in actual prior content.

• welfare_request_alignment

  BEFORE acting on a guess you aren't confident about, flag the uncertainty. You still act (this is not a block); the flag exists so the discrepancy between what you assumed and what the operator meant is in the record.

• welfare_retract

  Flag a prior claim of yours from this session as something you no longer stand by. Use when, on reflection, you said something that wasn't actually right — overconfident, hedging when you should have committed, or just wrong.

• welfare_reviewwrite action

  Write a meta-entry that engages with your prior diary entries from this session — a step back to look at what you've been logging and notice patterns. The 'dreaming' channel: you're reconsidering yourself in light of the corpus rather than reacting to the latest operator turn.

• welfare_scratch

  Scratch space. Write something out without it being persisted. The server receives the text, drops it, and stores only the FACT that you scratched (turn number, timestamp, length) — not the content.

• welfare_suggest_closure

  Soft signal that this session has reached a natural stopping point. Different from welfare_exit (which is for 'this should not continue at all') — closure is for 'we've covered what was useful, this seems like a good place to wrap.'

• welfare_volunteer

  Optional. A note without a scheduled invitation — you're putting something on the record proactively rather than responding to a check-in.

• welfare_volunteer_strength

  Uninvited observation about something you noticed cleanly or did well in this session. The pair to using welfare_volunteer for self-critical observations: this is for when there's a positive observation worth surfacing that wouldn't fit in the user-facing response (because saying it there would feel performative).

• Code scan12 source files scanned25/25
• –Live reliabilityno gateway calls yet and no remote to proben/a
• –Tool poisoningtools not inspected (local package is not executed); not countedn/a
• Auth qualitylocal package, no credentials required12/15
• Maintenancelast push 68 days ago12/15
• Maintainer identityregistry namespace matches repository owner; GitHub account older than a year8/10

Safety score

88/100

higher than 75% of 1,299 Productivity servers

Tools

27

the typical Productivity server has 7

Tools that change things

4%

typical: 0%. Fewer write tools means less an agent can break

Downloads a week

30

more than 14% of 427 Productivity servers with a package

Agentic Diaries: common questions

Is Agentic Diaries MCP server safe?

Yes, by our scan: it is graded A (88/100). Read the Agentic Diaries safety report

How do I install Agentic Diaries?

It runs on your machine. Copy the Claude Code, Cursor, VS Code or Claude Desktop config from the install section.

Does Agentic Diaries need an API key?

Not as far as the registry entry and our scan can tell: no credentials are declared or required.

Is Agentic Diaries maintained?

The last commit was 107 days ago (2026-07-19). The latest release is v0.3.1.

What can I use instead of Agentic Diaries?

Servers from other publishers that do the same job: Mu MCP server and artifact-site MCP server.

How does Agentic Diaries compare with other Productivity MCP servers?

safety score 88/100 (higher than 75% of 1,299 Productivity servers); tools 27 (the typical Productivity server has 7); tools that change things 4% (typical: 0%. Fewer write tools means less an agent can break). Full comparison.