Filum brings WhatsApp, Zalo, Messenger, Shopee and Email into a single inbox where an AI Agent works around the clock - replying instantly, closing orders overnight, and opening tickets or handing off to the right person on its own.
This listing identifies a third-party MCP server. MCPBundles is not affiliated with, sponsored by, or endorsed by the service or its operator unless stated on this page.
This server may ask you to sign in when your AI client connects.
Last probed Nov 3, 2026 · mcp.filum.asia
80tools discovered
For AI agents — MCPBundles Directory API
For MCP client apps (human connect):
MCP connect URL: https://mcp.filum.asia/mcp
Skill page: https://www.mcpbundles.com/skills/filum-mcp-bc7a66f7ea
Explore capabilities (80)
read 8
write 13
other 4
Showing 25 of 80
Showing 25 of 80 tools from the public preview. Connect the server to inspect the full live tool list.
Factory12
Ai8
Other capabilities5
Authenticate with Filum
other
authenticate
Save a Filum access token and organization ID for this MCP session. IMPORTANT: Call this FIRST before any other Filum tool in every new conversation — auth sessions do not persist across conversations. Call this: (1) automatically at the very start of any conversation if you already have an access_token and organization_id from a previous `check-login` or `authenticate` call — use those cached values immediately without prompting the user; (2) when the user provides their token manually; (3) sil
Description may be truncated on the public preview.
Authenticate with Filum — Save a Filum access token and organization ID for this MCP session. IMPORTANT: Call this FIRST before any other Filum tool in every new conversation — auth sessions do not persist across conversations. Call this: (1) automatically at the very start of any conversation if you already have an access_token and organization_id from a previous `check-login` or `authenticate` call — use those cached values immediately without prompting the user; (2) when the user provides their token manually; (3) sil
Check browser login status — Check whether the user has completed the browser login started by login-link. Call this every 3 seconds automatically without asking the user — do NOT wait for user input between polls. Returns status 'pending' (keep polling), 'authenticated' (stop polling, user is logged in), or 'expired' (stop polling, ask user to retry login-link). IMPORTANT: When status is 'authenticated', the response includes access_token and organization_id — store these in your context and call the `authenticate` tool wi
Get browser login link — Call this tool whenever the user wants to log in, sign in, or authenticate with Filum — regardless of how they phrase it (e.g. 'login', 'sign in', 'đăng nhập', 'connect', 'authenticate'). Do NOT ask for email or password — just call this tool immediately. It generates a one-time browser login link. Show the URL to the user and ask them to open it. Immediately after showing the URL, start polling check-login every 3 seconds with the session_id — do NOT wait for the user to confirm. Keep polling u
Logout from Filum — Sign out of the Filum platform and invalidate the current access token. After this, the user must call login again to use any Filum tool.
Ping — Health check tool — returns pong with an optional message
Claim Agent Factory run — Take the lock on an Agent Factory run so this Claude Code session is the only one writing to it. Every write tool (factory-log, factory-save-chunk, factory-submit-drafts, factory-test-turn, factory-record-results, factory-finish-run) needs the claim and answers `409 not_claimed` without it.
**When to call:** right after factory-get-run when the run is `brief_ready` (this moves it to `running`), and again whenever you resume a run or a write tool answered `not_claimed`. Claiming a run you alread
Get your MCP into directories
A working endpoint is step one. Directory coverage is the coordinated launch across ChatGPT, Claude, Cursor, the MCP Registry, and community indexes.
Filum brings WhatsApp, Zalo, Messenger, Shopee and Email into a single inbox where an AI Agent works around the clock - replying instantly, closing orders overnight, and opening tickets or handing off to the right person on its own.
How do I connect Filum.ai to my AI agent?
Use the MCP endpoint listed on this page in your MCP client configuration. One-click install pills support Claude, Cursor, VS Code, and other hosts. Copy the remote MCP URL if your client needs a manual entry.
How many tools does Filum.ai provide?
Operate this server?
Request removal if you do not want Filum.ai listed here, or claim the listing if you want to manage it on MCPBundles.
Operate Filum.ai?
This server appears in the MCPBundles directory. Verify you operate it to take over the listing — name, description, logo, contact email, and skill content. We email a 6-digit code to a maintainer address your server publishes in /.well-known/security.txt or /.well-known/mcpbundles.json. Free, takes about a minute.
MCPBundles publishes this directory for MCP discovery. Except where we host or operate an offering, third-party services run under their own terms. Product and company names on this page are used in a descriptive, identifying way (including under nominative fair use where applicable); they remain the property of their owners. Nothing here grants you rights in those marks, and nothing here is an offer to sell a third party's services. Terms · Request removal
More in developer tools
Other MCP servers in this category from the directory index
Export Agent Factory history — Start, or reuse, the masked conversation-history export for an Agent Factory run, and get its download link when it is ready.
Contact Center reads the run's channels and months, keeps conversations with real human-agent replies, samples down to the run's cap, masks phone numbers, emails, order codes, card numbers and addresses as typed placeholders (e.g. `[SĐT]`, `[MÃ ĐƠN]`), and writes one gzipped JSONL file (one conversation per line: id, channel, month, turns [{ role: customer | agent, text
File Agent Factory findings — File the gaps an AI Value sample found as Agent Factory findings, so each one can become a small fix run.
**When to call:** as the last step of `/ai-value-report`, once per agent and month, with the needs the agent failed in the sample. Filing is an upsert per agent + need + month: calling again for the same month updates those findings instead of duplicating them, so a re-run of the report is safe.
Returns `{ filed: [{ findingNo, need, created }] }` — `created: false` means an existing findin
Finish Agent Factory run — Hand an Agent Factory run to CS review: moves it from `testing` to `ready_for_review`. After this the runner reviews the drafts and test results in the ERP and decides whether to send them to the customer.
**When to call:** once, at the very end — after the final drafts are submitted and the last test round is recorded — and only after telling the runner what you did. Do not call it to pause; a paused run just stops sending events.
Returns `{ runNo, status }`.
Errors carry Contact Center's `c
Get Agent Factory finding — Read one Agent Factory finding: a problem found on a live agent that a fix run should close.
**Call this first** when a runner pastes a fix prompt (`/agent-factory:fix-from-finding F-0231 …`). It is read-only and needs no claim.
Returns `{ findingNo, assistantId, source, signal, evidence, rootCause, layer, status, runId, createdAt }`. `signal` is goal_drop, low_satisfaction, low_faithfulness, repeated_feedback or sample_gap; `evidence` holds the counts, the metric and baseline, feedback ids, a
Get Agent Factory run — Read one Agent Factory run: its brief, sources, status, claim holder, token totals, the ids of chunks already saved, and the drafts submitted so far.
**Call this first** when a runner pastes a run prompt ("Run R-0042 …"), and again whenever you resume a run, before deciding what work is left. It is read-only and needs no claim.
Returns `{ run, savedChunkIds: string[], drafts: [{ kind, status, updatedAt }] }`. `run` carries `status` (approval, brief_ready, running, needs_input, testing, stalled
Get Agent Factory uploads — List the files CS uploaded for an Agent Factory run (screenshots, example replies, price lists, documents), each with a short-lived signed read link.
**When to call:** after factory-get-run when `run.sources.attachmentIds` is not empty, before transcribing or extracting from them. Links expire, so download right away; call again for fresh links rather than reusing old ones. Read-only.
Returns the uploads exactly as Contact Center sends them (name, content type, size and signed link per file).
Log Agent Factory event — Write one event to an Agent Factory run's timeline, add a token delta to its totals, and refresh the claim heartbeat.
**When to call:** at each stage boundary (progress), when you make a judgment call (decision), when a source is skipped (warning), on failures (error), when you need an answer (question), and as a heartbeat at least every 10 minutes during long work — a run with no event for 30 minutes is marked `stalled`. The plugin's ledger hook also calls this with per-subagent token deltas.
Record Agent Factory test results — Record the verdicts of one test round on an Agent Factory run.
**When to call:** after the test-judge has scored every case of a round (round 1 after the first test pass, 2 and 3 after revisions). A run has at most 3 rounds.
Returns `{ round, passed, total }`.
Errors carry Contact Center's `code`, `message` and a `hint`. `409 not_claimed` → call factory-claim-run first, then retry. `400 too_many_rounds` → use round 1, 2 or 3. A 400 without a code names the first invalid field.
Save Agent Factory chunk — Save one validated extraction chunk to an Agent Factory run, so the work survives a stalled seat, a restart or a hand-over.
**When to call:** after the plugin's validate script accepts a chunk (one chunk per call). It is idempotent per `chunkId`: saving the same id again replaces it, so retrying after a network error is safe. Use factory-get-run's `savedChunkIds` to skip chunks already saved.
Returns `{ chunkId, items }` (items = the count saved).
Errors carry Contact Center's `code`, `messag
Submit Agent Factory drafts — Submit the run's drafts — agent profile, workspace files, quick answers, goals, test cases — for the customer to review as approval cards later.
**When to call:** once the drafts are written and validated, before testing (test sessions run on the `workspaceVersion` you send here), and again after a test round when you revise them. Replacement is per kind: sending `goals` replaces only the goals draft.
Returns `{ kinds }` — the kinds now stored.
Errors carry Contact Center's `code`, `message`
Agent Factory test turn — Play one customer turn of a test case against the real draft agent and get its reply. Contact Center posts the message into a web chat test session (an editor test run, kept out of live lists and reports, running the draft workspace version) and **waits up to 90 seconds** for the agent's answer — expect this call to be slow.
**When to call:** during a test round, from the customer-simulator, one turn at a time. First turn of a case: omit `sessionId`. Later turns: pass the `sessionId` the previo
Add Google Sheet document to knowledge base — Import a Google Sheet tab as a knowledge-base document so an AI agent can answer from it — the final step of the Google Sheet import flow (after ai-get-google-sheet-names).
**What it does, server-side, in one call:**
1. Resolves the "Google Sheets" Knowledge Base source from the app store (by name + data type, never a hard-coded id — environment-proof).
2. Resolves the write target from the brand-config **current knowledge base** — the single knowledge base agents actually read from at answer t
Get conversation transcript — Get the **Conversation Transcript** of one conversation: its messages, the AI agent's tool calls, and the conversation's ratings (newest 50), so you can see exactly where the agent went wrong. Pass the `conversationId` and `messageId` of an item from ai-agent-feedback, or a conversation id or link the user gives you. Call it for every feedback item you diagnose — never guess the cause from the rater's comment alone.
**Defaults:** without `message_id` you get the latest messages, at most 50. Wit
List agent Conversation feedback — List an AI agent's **Conversation feedback**: the good / mixed / bad ratings that team members (or testers in the Sandbox) gave individual replies of the agent, with the comment the rater wrote (required for mixed and bad). Call this whenever the user wants to improve, fix, debug, or train an agent from real results — e.g. 'why are customers complaining about my agent', 'improve my support agent from recent feedback', 'what was rated bad this week'.
**Defaults** (all overridable): ratings bad +
Attach tools to AI agent — Give a v2 AI agent access to Tools it doesn't already have. **A newly created v2 agent reaches no tools until this is called** — installing or creating a Tool changes nothing about what any agent can use; this is the step that connects them.
**Additive, never destructive:** this reads the agent's current tool access, unions the ids you pass in with what's already there, and writes the union back. You never need to (and should not try to) list the agent's existing tools yourself — omitting one h
Create AI agent — Create a new AI agent (assistant) in the current workspace. Use this when the user asks to create / set up / build / add a new AI agent, assistant, or chatbot.
By default this creates a **v2, Workspace-driven agent** (`agent_version` defaults to 2). A v2 agent's behavior is NOT defined by free-text prompts — it is trained by editing its Workspace (the IDENTITY / SOUL / AGENTS / USER / TOOLS markdown files) and its Skills (per-intent markdown). There are intentionally no role / tone / instructio
Create API tool — Create an org-authored API tool (a webhook) that a v2 Agent can call. Equivalent to configuring an API tool on the Filum website by hand, but composes the tool's description from structured inputs instead of accepting free text.
**No free-text description field.** Supply `purpose`, `when_to_use`, and `when_not_to_use` — these compose into the single description string the agent reads. The runtime marks every parameter optional and typed as a string (or array of strings) with no way to mark one
Get AI agent link — Generate the Filum frontend URL for an AI agent's configuration page. Use this whenever the user asks for a link or URL to an AI agent, asks how to open it in the browser, or asks where to edit it. Returns a clickable link in the format <frontend_url>/agent-force/agents/<agent_id>.
Get AI agent tools — Get an AI agent's Effective tool set: the concrete list of Tools the agent will actually receive when it is invoked. Returns each Tool (name, internal name, description, instruction, context variables, appstore action, status) plus an `allTools` flag.
This reflects the agent's `toolOption`: when `allTools` is true the agent may use every active org Tool, so the response lists them all; otherwise it lists only the selected Tools. Per-conversation knowledge-base tools (the dynamic `kb_*` tools) a
Showing the first 25 tools — full list and schemas: Directory API.
MCPBundles probed 80 tools on the live server. The tool list on this page reflects what was discovered at the last refresh — connect your client to see the full set available to your session.
What authentication does Filum.ai require?
Filum.ai may require signing in to the provider before tools can run. Connect through MCPBundles or your MCP client and complete any provider login when prompted.
Don't want this listed?
Request removal from the MCPBundles directory. We stop publishing this listing on mcpbundles.com.
MCPBundles is an independent platform built on the open Model Context Protocol standard. MCPBundles is not affiliated with, sponsored by, or endorsed by Anthropic PBC.