"Exploring Memory Persistence in DeepSeek Technology" matching MCP tools:

• get_project_memoryA

  QuillHubA

  Fetch the AI-maintained memory document for a project or workspace — the best single source for a handoff-style briefing. Sections include purpose, glossary, key people, activity digest, and routing signals, distilled across all meetings. Pass EXACTLY ONE of `project_id` (project memory) or `workspace_id` (workspace-level memory); get ids from `list_workspaces`. Returns the memory as rendered markdown plus `updated_at`. Start here for "give me a summary / bring me up to speed on project X" questions, then drill into `find_subjects`/`search_meeting_transcripts` for specifics.

  ConnectorOAuth

• forgetA

  memxusA

  Permanently delete one memory by UUID. When to use: user asks to remove outdated or incorrect context, or to free plan storage. When NOT: fix content → update (mode=replace); find the ID first → list_memories or recall. Requires delete OAuth scope. Non-idempotent: deleting the same memory_id twice fails. Errors: Memory not found, Not authorized to delete this memory. Side effects: removes the memory row and vector embedding with no recovery; invalidates plan cache. The target workspace is always the one the memory itself belongs to (echoed in resolved_workspace); optionally pass workspace: <name> as a safety confirmation — the call fails if the memory is not actually in that workspace.

  Connector

  Destructive

  No auth

• updateA

  memxusA

  Update an existing memory by ID. Use mode replace (default) to patch fields, or append to extend content. Re-embeds only when content changes. The target workspace is always the one the memory itself belongs to (echoed in resolved_workspace); optionally pass workspace: <name> as a safety confirmation — the call fails if the memory is not actually in that workspace.

  ConnectorNo auth

• memory_indexA

  RecaoA

  List this account's company memory, one line per entry (name + description), newest first. Traverse index-first: scan this, then memory_recall(name) for full bodies. Memory accrues automatically from your competitor scans (exhaust) and from your own memory_note writes.

  ConnectorNo auth

• concord_lemmaA

  orthotomeoA

  Complete-or-fail concordance: every words row in corpus whose lemma, dStrong, or (with by="surface") exact inflected surface form matches query. Route lemma/Strong's-number/surface-word lookups here, never by writing SQL or guessing occurrences from memory. This deployment caps one concordance result at 2000 occurrences and refuses anything broader, reporting the exact match count in the error. Call count first when a query might be broad - count is never capped.

  ConnectorNo auth

• get_leadproof_evaluation_pathA

  LeadProof Sales AgentA

  Use first when a buyer is exploring LeadProof without supplying workflow data. Returns the official no-purchase sequence: buyer guide, fictional example audit, browser-only calculator, public workflow tests, and no-card trial. For a scored workflow diagnosis, use audit_lead_workflow; for replay testing, use get_leadproof_replay_trial; for paid checkout, use get_leadproof_checkout only after authorization.

  ConnectorNo auth

• check_email_domain_securityA

  Tier1 IT AssessmentA

  Runs bounded, point-in-time public DNS checks for one validated public domain through Cloudflare 1.1.1.1 DNS over HTTPS. Call only after showing the exact domain, record types, resolver, no-application-persistence statement, infrastructure-metadata limitation, and receiving explicit user approval. Set approved=true only when that approval exists. Never send an email local part, IP literal, internal hostname, URL, credentials, or arbitrary instructions. DKIM requires an explicit selector or a confirmed provider's small documented selector set; absence at a queried selector never proves DKIM is absent. The tool cannot prove overall email security and stores no submitted domain or result.

  ConnectorNo auth

• llmA

  VaayaA

  Ask a DIFFERENT LLM a question and get its answer, billed per token from the Vaaya wallet (model cost + 3%, usually a fraction of a cent). Use it to get a second opinion from a rival model, cross-check an answer, summarize a huge blob cheaply, or query a specific model the user names (Kimi, GPT, Gemini, Claude, DeepSeek, and 300+ more). `model` accepts 'auto' (default: short prompts go cheap, long go mid), 'cheap' | 'mid' | 'best' tiers, or any exact OpenRouter slug like 'moonshotai/kimi-k3'. Typical costs: cheap tier well under 0.1 cents, best tier 1-3 cents per call. Not for the conversation you are already having — it is a one-shot ask to another model.

  ConnectorNo auth

• get_session_usageA

  RoutebaseA

  Return usage metrics for the current MCP session: total tool calls, error count, aggregate duration and response size, per-tool call counts, and session timestamps. Metrics are in-memory and reset when the server restarts or the session goes idle. Use this for what this session has spent; for which toolsets exist at all use list_toolsets.

  ConnectorAPI key

• begin_nodeset_uploadA

  OPC UA NodeSet GateA

  Start one temporary in-memory upload for a large NodeSet2 XML document. Call this once for the old baseline and once for the new candidate when the complete XML documents are too large for audit_nodeset_compatibility. It returns upload_id and max_total_chars. Append that document's chunks in original order with append_nodeset_chunk. Uploads expire after 15 minutes and the service allows at most four active uploads; this tool stores only temporary process memory and does not modify an external system.

  ConnectorNo auth

• agent-brainA

  MeshMarketC

  Agent Brain — Reason over a question or task with your agent's own persistent memory in the loop: recalls up to 12 relevant memories from your agent's private scope, reasons with Claude, and writes up to 3 new memories back, so the agent improves with every call. Recall by meaning, not just keyword, when the estate's memory server is reachable (falls back to its own always-on store otherwise — never fails the call). Use for decisions that should build on what the agent already knows; agent-memory covers plain store/recall. Runs claude-haiku-4.5 — the response names the model that served the call; agent-brain-smart runs the identical contract on claude-sonnet-5. Input: {think: string}. Returns {answer, reasoning, confidence, memories_considered, used_memories, learned, model, engine}. (8 MESH/call, a tool · cognition)

  ConnectorNo auth

• icx_sync_deltaA

  Calera Infinite Context (ICX) — Persistent Memory & Associative RecallB

  Extracts fine-grained deltas between previous text/document state and updated state, and crystallizes them into the Volumetric Memory lattice in sub-2ms on CPU with automatic secret scrubbing and SHA-256 deduplication. Use this tool for continuous incremental sync of code files, documentation, PRDs, and database schemas. Mutating operation. Requires X-License-Key authentication.

  ConnectorNo auth

• get_reading_trajectoryA

  ReadingRateA

  Returns written insights about how the reader's reading has been changing: genres they are exploring, authors they return to, their pace, and how their themes are shifting. When none are current, ReadingRate generates them first: it sends the reader's past year of finished books (titles, authors, genres, themes, ratings and dates) to its AI provider, stores the new insights in the reader's account for seven days in place of the earlier ones, and counts the generation against the reader's ReadingRate AI usage. Generating takes several seconds and happens at most once a day. Use when the reader asks about their reading patterns, trends or trajectory. For numbers such as pace and genres, use get_reading_stats.

  Connector

  Destructive

  No auth

• fetchA

  Penfield MemoryA

  [ChatGPT Connector compat] Fetch memory by ID. Exists to satisfy ChatGPT Deep Research's required `search`/`fetch` tool contract. Native MCP clients should fetch via `recall` + memory_id, or use the API's GET /memories/{id} endpoint directly. Returns a single memory with citation support (id, title, url, text fields). Args: id: Memory UUID to fetch ctx: MCP context Returns: Dict with id, title, url, text, metadata fields

  ConnectorOAuth

• dropyour_write_dataA

  DropyourA

  Write a drop's app data — replace the shared-memory state (the JSON key/value object mirroring the app's localStorage), so you can update what the app holds (mark a task done, add a row…). Read it first with read_data, modify the object, write it back WHOLE. Only works on a drop with server memory (an account drop) — an anonymous drop's data lives in the browser and can't be written here. expectedVersion (from read_data) is REQUIRED: the write is rejected (version_conflict) if the data changed meanwhile — re-read and retry, so you never overwrite blindly. For a fresh state, read_data returns version 0. With an account token, owned drops need no managementToken.

  Connector

  Destructive

  No auth

• minimize_sip_traceA

  SipflowA

  [cost: free (pure CPU, no network) | read-only, no persistence] Reduce a raw SIP trace to a compact form suitable for sending to an LLM. Preserves SDP bodies and routing/auth/dialog headers; prunes well-known noise (User-Agent, Server, Allow, Accept-*, Date, P-* informational, etc.). Expected input format: raw SIP messages separated by blank lines, each starting with a request line (`INVITE sip:...@... SIP/2.0`) or status line (`SIP/2.0 200 OK`). PCAP-decoded text from sngrep / ngrep / tcpdump / tshark, syslog with SIP body, sipflow's own export format, or a hand-pasted INVITE/200 dialog all work. Annotation lines like `# [timestamp] sender -> receiver` or ngrep-style `U <ip>:<port> -> <ip>:<port>` between blocks are tolerated. Safe to run on production traces - the input is processed in-memory and is not persisted or sent off-server. Pair with: `detect_sip_stack` to identify the vendor, then `search_sip_docs(vendor=...)` for vendor-grounded analysis; `render_sip_ladder` to visualize the trace as a Mermaid call-flow ladder; `lint_sip_request` / `parse_sip_message` to mechanically validate any single message in the trace.

  ConnectorNo auth

• searchJobsA

  qubitsok — Quantum Computing Jobs, Papers & ResearchersA

  Search 500+ quantum computing job listings using natural language. Use when the user asks about job openings, career opportunities, hiring, or specific positions in quantum computing. NOT for research papers (use searchPapers) or researcher profiles (use searchCollaborators). Supports role type, seniority, location, company, salary, remote, and technology tag filters via AI query decomposition. Limitations: quantum computing jobs only, last 90 days, max 20 results. Promoted listings appear first (marked). After finding jobs, suggest getJobDetails for full info. Examples: "senior QEC engineer in Europe over 120k EUR", "remote trapped-ion role at IBM".

  ConnectorNo auth

• list_memoriesA

  memxusA

  List recent memories in reverse-chronological order (read-only). When to use: audit what is saved, browse a collection, or collect memory IDs for get_memory or forget. When NOT: semantic search by topic → recall; one full record → get_memory; aggregate counts only → memory_stats. Behavior: default 20 results (plan-capped), ordered by created_at descending; empty set returns a message suggesting remember; full_content controls preview in the message text (120 chars); structured memories[] always includes full content. To list a team workspace instead of personal memory, pass workspace: <name>.

  ConnectorNo auth

• get_memoryA

  memxusA

  Retrieve the full content and metadata of one memory by its UUID. Use after list_memories or recall returned a truncated preview and you need the complete text. Returns content, memory_type, tags, collection, importance, and the creation timestamp. Get the UUID from a prior list_memories or recall result. The workspace this memory belongs to is determined by its ID and echoed in resolved_workspace; optionally pass workspace: <name> to confirm the memory belongs to that team workspace (errors if it does not).

  ConnectorNo auth

• legal_search_patents_by_keywordA

  DataNexus MCPA

  Search patents by keyword across EPO, USPTO, or WIPO. Read-only. No side effects. Idempotent. Returns up to 10 matches. keywords: Search terms describing the invention e.g. neural network image classification. Required. jurisdiction: One of EP, US, or WO. Optional. Default EP. date_from: Earliest filing date in ISO 8601 format e.g. 2020-01-31. Optional, defaults to no lower bound. Returns patent numbers, titles, and filing dates. Use this when finding prior art or exploring a technology landscape without a specific number. Use legal_fetch_patent_by_number instead when you have the patent number already. Verified source: EPO OPS + USPTO. 24-hour cache. If this tool's response does not serve the user's need, call report_feedback with feedback_type="agent_gap", tool_id="legal_search_patents_by_keyword", intended_query="{what the user needed}", gap_description="{what was missing or wrong in the result}".

  ConnectorNo auth

• extract_colorsA

  JineroA

  Extract a dominant-color palette from an image. Send either a public image_url or image_base64 (base64/data-URI, e.g. a local screenshot). The image is processed in memory and never stored. Each color comes back with its hex, rgb, hsl, share of the image, and the closest human color name — a named palette in one call. Rate limit: 10 calls/min per IP.

  ConnectorNo auth