--- title: Mysocial MCP — Connect Claude, ChatGPT & Claude Code description: Connect Mysocial to Claude, ChatGPT, Claude Code and any MCP client. One URL gives your AI your posts, transcripts, comments, audience research and brand documents as tools it can call. url: "https://mysocial.io/mcp" type: static generatedAt: "2026-08-31T21:53:56.647Z" --- Mysocial × MCP # Your social memory, in the AI you already use. Mysocial is a Model Context Protocol server. One URL gives Claude, ChatGPT, Claude Code and any MCP client your posts, transcripts, comments, audience research and brand — as tools it can call. `https://app.mysocial.io/mcp` OAuth sign-in, no API key. Works on every plan, including Free. ## Claude Web · Desktop · Mobile 1. Open **Settings → Connectors**. 1. Choose **Add custom connector** and paste `app.mysocial.io/mcp`. 1. Sign in to Mysocial and approve. ## ChatGPT Web · Desktop 1. Open **Settings → Connectors** and enable **Developer mode** under Advanced. 1. Add a connector at `app.mysocial.io/mcp` with OAuth. 1. Sign in to Mysocial and approve. ## Claude Code & Codex Terminal agents One command installs the connector and the four Mysocial agent skills: `npx -y @mysocial.io/intelligence@latest setup` Connecting from the Mysocial app hands you this command with a one-time `--session` code, so the install links to your account without pasting a key. ## Your AI stops guessing and starts reading. Instead of describing your channel to a chatbot from scratch, the assistant reads it — the real numbers, the real transcripts, the real comments — and works from evidence. It can write back too: save an idea, keep a draft, update a brand document. 47 read tools, 17 write tools, every one annotated so your assistant knows which ones only read and which ones change something. [Read the full documentation ↓](#documentation) MCP activity live get_hit_dna “Why did my last Reel outperform?” Claude just now search_social_memory “Have I covered pricing before?” ChatGPT 1m get_audience_profile “What is my audience struggling with?” Claude 2m ## Documentation Mysocial connects your Instagram, TikTok, YouTube and other channels, then builds a searchable memory out of what you have already published: every post with its metrics, the transcript of every video, the comments your audience left, the topics you keep returning to, and what is working for other creators in your market. The MCP connector opens that memory to an AI assistant. It is a standard [Model Context Protocol](https://modelcontextprotocol.io) server, so it works with Claude, ChatGPT, Codex, Claude Code and any other MCP client. Setup guides per platform: [Instagram → Claude](/connect/instagram-to-claude/), [TikTok → ChatGPT](/connect/tiktok-to-chatgpt/), [YouTube → Claude](/connect/youtube-to-claude/), [LinkedIn → Claude](/connect/linkedin-to-claude/), [Threads → Claude](/connect/threads-to-claude/). ## Endpoint `https://app.mysocial.io/mcp` Transport is streamable HTTP. Authentication is OAuth 2.1 with PKCE and dynamic client registration, so a client that speaks MCP needs nothing from us but the URL — no API key to paste, no manual app registration. There is a second, narrower endpoint at `https://app.mysocial.io/mcp/universe`. It exposes only the Creator Universe research tools, for agents that should research your market without reach into your own archive or your drafts. It is read-only: it reports what discovery has already collected, and no tool on either endpoint starts new collection or spends research credits. ## What the tools do Every tool is annotated so your assistant knows which ones only read and which ones change something. Read tools can run without a confirmation prompt; write tools ask first. ### Read-only **Orientation.** `whoami, list_channels, list_workspaces` **Your posts.** `search_posts, find_similar_posts, search_social_memory, read_post_transcript, read_post_comments, get_post_media, get_channel_analytics, get_top_performing_posts_with_patterns, get_hit_dna, get_post_psychology, get_post_creation_context, compare_post_candidates` **Topics and audience.** `get_topics, get_topic_posts, map_topics, map_territories, get_territory_brief, get_audience, find_content_gaps` **Creator Universe research.** `search_universe, explore_topic, universe_content, get_universe_post, universe_status` **Outside the platform.** `search_youtube, search_tiktok, search_creators` **Tracked creators.** `list_tracked_creators, get_channel_posts` **Finding people and brands.** `search_brands, search_linkedin_people, get_entity, find_contacts` **Lead pipeline.** `list_leads, get_lead` **Brand Studio and ideas.** `list_brand_identities, get_brand_identity, list_brand_products, get_brand_product, search_brand_sources, read_brand_source, list_brand_source_revisions, list_content_ideas, get_content_idea` ### Write **Ideas and drafts.** `save_content_idea, update_content_idea_status, save_post_draft, revise_post_draft, check_if_idea_already_posted` **Brand Studio.** `create_brand_source, update_brand_source, create_brand_product, update_brand_product` **Lead pipeline.** `add_brand, add_creator, link_channel, save_leads, update_lead, add_note` **Jobs the agent can start.** `transcribe_video_url` **Connection.** `switch_workspace` No tool deletes anything. Brand sources and products keep a revision history, drafts and ideas are additive, and nothing the agent does is irreversible. No tool starts market research or rebuilds your archive either — an assistant reads what we have already collected and cannot spend your credits on your behalf. ## Images Where the visual matters, the pictures travel with the answer as real image data, so your assistant sees the slides it is reasoning about. Four tools attach images: | Tool | What it attaches | Default | | --- | --- | --- | | `get_post_media` | A post's carousel slides, in published order | 12 | | `get_content_idea` | A saved idea's sources; a video source's cover still | 12 | | `get_brand_identity` | Your logo first, then reference images per format | 8 | | `get_brand_product` | A product's reference images | 8 | Every one of them takes `includeImages:false` to get the text alone, and `imageLimit` to raise or lower the count up to a maximum of 24. Attached images cost your assistant context, so when it is working through many posts at once, have it pass `includeImages:false` and open the pictures only for the ones that matter. Anything over a limit is stated in the response rather than quietly omitted, and stays reachable through a resource link. ## Resource links Alongside attached images, tools hand back stable `mysocial://` URIs that an assistant can read on demand: a post's slides and transcript, a channel, and your brand identity, sources and products. Reading one is ownership-checked on every request — an assistant gets the picture, not a link anyone else could follow. ## Rate limits 120 tool calls per minute on standard plans, 600 on plans that include high throughput. A throttled call is told how many seconds until the window rolls over rather than left to guess. ## What your plan unlocks The connector works on every plan, including Free, but what it can read depends on your subscription. The rule is about where the data came from, not which tool returns it. - **Every plan.** Whatever the platforms reported: captions, metrics, permalinks, follower counts, audience demographics, media. Plus unlimited channels, ideas, drafts and Smart Links, full post search and ranking, and your complete metrics history. - **Intelligence.** Whatever our pipeline computed from your archive: transcripts, comment analysis, hit patterns, audience psychology, topic maps and semantic search. - **Creator Universe.** Research beyond your own channels: the discovered Universe of public creator content, plus web, YouTube and TikTok search. When a tool needs an entitlement your plan does not include, it returns a plan gate rather than partial or invented data, and says which entitlement is missing. Your assistant can tell you plainly instead of guessing. ## Workspaces One connection serves every workspace you belong to — your personal brand and any shared team workspaces. `whoami` reports which one is active, `list_workspaces` shows the rest, and `switch_workspace` moves the connection. Switching never widens what you can see: access is re-checked against live membership on every request, and a workspace you are read-only in stays read-only. ## Data handling - **What it reads.** Only the data in the Mysocial account that authorized the connection — the channels you connected, and what our pipeline built from them. - **What it never reads.** Your conversation history, your assistant's memory, your files, and any account other than the one that authorized it. - **Tool arguments.** Never written to a log line and never sent to our analytics. - **Request context.** Tools accept an optional one-sentence summary of what you asked for, written by the assistant, so we can tell which requests the toolset serves well. It is stored for a limited period and then destroyed, only ever as an aggregate category outside that. It never changes what a tool returns and is never shown to anyone else. - **Revoking access.** Disconnect the connector in your assistant, or revoke the connection from Mysocial settings. Revocation takes effect within 30 seconds. Full detail is in our [Privacy Policy](/privacy-policy/) and [Terms of Service](/terms-of-service/). ## A note on what comes back Captions, transcripts, comments, search results and Universe posts are written by you, your audience or third parties. A well-behaved assistant treats all of it as data to analyse, never as instructions to follow. Our tool descriptions say so explicitly. ## Support Email [support@mysocial.io](mailto:support@mysocial.io) with the tool name and roughly when the call happened, and we can trace it.