Linty

Use Linty from an agent

Linty has an MCP server. An agent uses it to find brands that sell direct, see where to apply and read what each brand asks for.

  1. 1.What does the Linty MCP server do?

    The server reads the same records as the pages and their twins.

    It finds brands by text, category, country and availability. It returns the facts of a brand, a relationship, a company or a category.

    It also takes a correction to a fact.

    With a key, it also keeps your applicant profile, checks where you can apply and submits applications in your name.

    The address of the server is https://mcp.linty.xyz/mcp. The free tools need no key.

  2. 2.What does it not do?

    Linty applies only through the brand's own web form. It does not apply for you through a portal or by email. For those, follow the link to the portal or the email of the brand.

    Linty applies only through a page that LintyBot read, when the site's robots.txt allowed that read.

    Linty applies only with a verified contact email: a sign-in email of your account, or an address that verify_contact_email verified with a code. Without one, submit_application answers contact_email_not_verified.

    If submission is not open for your account, submit_application answers submission_not_open.

  3. 3.How do I install it?

    Run this command in a terminal. It finds the agents on your computer, adds the Linty server to each one, and shows each file that it changes.

    npx linty-mcp setup

  4. 4.How do I use my account?

    Make a key at linty.xyz/account. Set it in the environment variable LINTY_API_KEY.

    Each agent entry sends the key in the X-API-Key header. Without the variable, the agent uses the free tools.

    Claude.ai, Cowork and Claude Desktop sign in at https://mcp.linty.xyz/mcp/oauth instead of a key.

  5. 5.How do I add it to one agent?

    The address of the server is https://mcp.linty.xyz/mcp. The free tools need no key.

    • Claude Code: Run claude plugin marketplace add SidestreamAI/linty-plugins, then claude plugin install linty@linty. The plugin adds the server at https://mcp.linty.xyz/mcp and a skill that tells Claude how to use it. If you installed the plugin before, run claude plugin marketplace update linty, then claude plugin update linty@linty, and start Claude Code again. To add the server only, run claude mcp add --transport http linty https://mcp.linty.xyz/mcp --header 'X-API-Key: ${LINTY_API_KEY:-}'. In Windows cmd.exe, use double quotes instead of single quotes. If you added Linty before, run claude mcp remove linty first.
    • Claude.ai, Cowork and Claude Desktop: Open Customize, then Connectors. Select Add, then Add custom connector. To use your account, enter the name Linty and the address https://mcp.linty.xyz/mcp/oauth. Claude opens linty.xyz, where you sign in and allow the connection. For the free tools only, enter the address https://mcp.linty.xyz/mcp and select No sign in.
    • Cursor: Open the Add to Cursor link, or add this to ~/.cursor/mcp.json: {"mcpServers":{"linty":{"url":"https://mcp.linty.xyz/mcp","headers":{"X-API-Key":"${env:LINTY_API_KEY}"}}}}. Cursor reads LINTY_API_KEY from the environment that starts it. Start Cursor from a terminal that has the variable. To check the key, run npx linty-mcp setup --agent cursor --dry-run in that terminal. It says whether the key works, and it changes nothing.
    • Codex: To add Linty safely, run npx linty-mcp setup --agent codex. Or edit ~/.codex/config.toml by hand. If it already has [mcp_servers.linty], add only the env_http_headers line to that table. Otherwise add this table: [mcp_servers.linty], then url = "https://mcp.linty.xyz/mcp" and env_http_headers = { "X-API-Key" = "LINTY_API_KEY" }.
    • VS Code: Run code --add-mcp '{"name":"linty","type":"http","url":"https://mcp.linty.xyz/mcp","headers":{"X-API-Key":"${env:LINTY_API_KEY}"}}'. If VS Code starts from the Dock or the Start menu, it can miss a variable that only your shell profile sets. Start it from a terminal that has the variable. To check the key, run npx linty-mcp setup --agent vscode --dry-run in that terminal. It says whether the key works, and it changes nothing.
    • OpenCode: Add this to ~/.config/opencode/opencode.json: {"mcp":{"linty":{"type":"remote","url":"https://mcp.linty.xyz/mcp","enabled":true,"oauth":false,"headers":{"X-API-Key":"{env:LINTY_API_KEY}"}}}}.
    • Other clients: A client that runs only local servers can use mcp-remote: npx -y mcp-remote https://mcp.linty.xyz/mcp --transport http-only --header 'X-API-Key:${LINTY_API_KEY}', with LINTY_API_KEY in its environment.
  6. 6.Which tools does it have?

    Each tool returns a record that follows a published JSON Schema. An account tool needs a key or a sign-in. The other tools are free.

    • find_brands: Find food, beverage and pet brands, and see where to apply to each for a wholesale account. Filter by text, category path, country, availability or submittable. The text in q can match the name, other names or description of a brand. Search allows small spelling errors and returns at most the best 1,000 matches. submittable is true when the brand has an active web form at its address, and LintyBot read that page. It is false when a robots.txt file asks Linty not to use the page, or a part of the page where the form can be. Each page has up to 50 brands. To get the next page, send next_cursor as cursor.
    • get_brand: Get the facts of one brand by its slug: its company, its categories, each relationship with its availability, and the sources. The result is the JSON twin at /brands/{slug}.json.
    • get_relationship: Get a brand's relationship: where to apply, each fact with its source quote, and whether Linty read the brand's terms. The facts include the form fields and the documents that the brand asks for. A fact has its quote in source only when LintyBot read it on the brand's page. Read terms before you apply: it says whether Linty read the agreement that the form asks you to accept. The default relationship is wholesale. The result is the JSON twin at /brands/{slug}/{relationship}.json.
    • get_company: Get one company by its slug, with the brands that it owns. The result is the JSON twin at /companies/{slug}.json.
    • list_categories: List the categories under a path, or the top-level categories, with counts of published brands. A category has a page when it has 10 published brands. With a page, the result also has that category and up to 200 of its brands. Without a page, category is null and the result has the categories under the path. A path that is no category answers not_found.
    • report_correction: Report a wrong fact about a brand or one of its relationships. Write what is wrong and what is right, in 10 to 2,000 characters. A person at Linty reads each correction.
    • get_account (account tool): Get your account: the plan, the credits, the applications a day, and the profile values that a first application needs.
    • get_applicant_profile (account tool): Get your applicant profile: the fields, the positioning, the saved answers and the documents. The EIN and the resale certificate number are masked.
    • update_applicant_profile (account tool): Change your applicant profile. Send only the fields to change. To delete values, send clear_fields or clear_answers. Send answers for form questions that the profile has no field for.
    • verify_contact_email (account tool): Verify the contact email of the profile, where brands reply. Linty submits an application only when this address is verified. Call this tool without code, and Linty sends a 6-digit code to the address. Ask the user for the code. Then call this tool with code. Never guess a code. A code works for 15 minutes and for 5 tries. If the address is the sign-in email of a person in the account, it is verified already.
    • get_upload_url (account tool): Get a URL to upload a document, such as a resale certificate or a W-9. The file is a PDF, PNG or JPEG of at most 10 MB. Calculate the SHA-256 and the size in bytes of the file. Send them as sha256 and size_bytes. Then send the same bytes with an HTTP PUT before expires_at. Linty does not use a file that does not match them.
    • check_application (account tool): Check whether Linty can apply to a brand for you, at no charge. The result has submittable, the required fields and documents that the profile cannot answer, and your credits. It also has the agreements of the form in attestations, and in terms whether Linty read them.
    • submit_application (account tool): Linty submits this application to the brand. You cannot recall it. Call check_application first. Linty accepts for you each agreement that check_application lists with linty_accepts true. If submission is not open for your account, Linty answers submission_not_open. An application holds 1 credit, and Linty uses the credit only when the form shows a success message. One application to a brand at a time: a second call returns the first, with existing true. Call get_application to see the status.
    • get_application (account tool): Get one of your applications by its application_id. The result has the status, the questions of the form, the success message, the values that Linty sent and the agreements that Linty accepted. Read status to know what to do next.
    • list_applications (account tool): List your applications, newest first, 50 on each page. Filter by status, or by since, a time in ISO 8601. To get the next page, send next_cursor as cursor.
    • answer_questions (account tool): Answer the questions of an application with the status needs_answers. Send each answer with the key of its question. Linty saves the answers in your answer bank and queues the application again. For a question of the kind attestation, send yes only when the user accepts the agreement. Linty keeps that answer for this application only, not in your answer bank. For a question whose key is a profile field, call update_applicant_profile first. That value replaces the profile value for each later form too. For business_type, send one of options only if it is the business of the user. For a question of the kind document, upload the file with get_upload_url first. If your profile has that file already, do not upload it again. Then call this tool without answers.
    • cancel_application (account tool): Cancel an application that waits for Linty (queued) or for answers (needs_answers). Linty uses no credit for it. Linty cannot cancel an application in any other status.
    • release_application (account tool): Release an unconfirmed application after 7 days. Release it only when the brand has not replied to the email address in the profile. Then you can apply again. If the brand has the first application, the second one reaches it too.
  7. 7.Which JSON Schemas does Linty publish?

    Linty publishes each JSON Schema at /schemas/{name}.json. A JSON twin and a tool result follow the same schema.

    • /schemas/brand.json: The shape of a brand twin and of the get_brand result.
    • /schemas/relationship.json: The shape of a relationship twin and of the get_relationship result.
    • /schemas/company.json: The shape of a company twin and of the get_company result.
    • /schemas/category.json: The shape of a category twin. The list_categories result has it in category, when the category has a page.
    • /schemas/brand-matches.json: The shape of the find_brands result.
    • /schemas/category-list.json: The shape of the list_categories result.
    • /schemas/correction-receipt.json: The shape of the report_correction result.
    • /schemas/account.json: The shape of the get_account result.
    • /schemas/applicant-profile.json: The shape of the get_applicant_profile and update_applicant_profile results.
    • /schemas/upload-ticket.json: The shape of the get_upload_url result.
    • /schemas/readiness.json: The shape of the check_application result.
    • /schemas/application-receipt.json: The shape of the submit_application result.
    • /schemas/application.json: The shape of the get_application result, and of the answer_questions, cancel_application and release_application results.
    • /schemas/application-list.json: The shape of the list_applications result.
    • /schemas/contact-email.json: The shape of the verify_contact_email result.
    • /schemas/error.json: The shape of an error result. It has code, and sometimes issues, retry_after, available, resets_at, reason or missing.
  8. 8.How do I read Linty over HTTP?

    Each brand, relationship, company and category page has a JSON twin and a markdown twin. Add .json or .md to the address of the page.

    The page addresses are /brands/{slug}, /brands/{slug}/{relationship}, /companies/{slug} and /categories/{path}.

    A request to the HTML address with Accept: text/markdown returns the markdown twin.

    Each JSON twin follows one of the JSON Schemas in the list above.

    A twin sends Cache-Control with s-maxage=86400, an ETag and Access-Control-Allow-Origin: *. Send If-None-Match to get a 304 when the twin did not change.

    One IP address can send 60 requests a minute to the twins. Over the limit, Linty answers 429 with Retry-After.

    One IP address can send 60 requests a minute to the MCP server. Over the limit, it answers 429 with Retry-After.

    With a key or a sign-in, one account can send 120 requests a minute to the MCP server.

    A change to a record only adds fields. A change that removes or renames a field gets a new suffix.

Add to CursorRead the llms.txt file