Add Mcp Integration logo

Add Mcp Integration

OrganizationPopular
vellum-ai
add-mcp-integration

Add a vendor's remote MCP server to the bundled integrations catalog (plugins/mcp-catalog/<name>/ + plugins/marketplace.json) so users can connect it from the Integrations page. Covers vetting the server's OAuth support, the icon, the package files, the generators, the CI checks, and the PR. Use when asked to "add <vendor> MCP", "add <vendor> to integrations", or to check whether a vendor's MCP server is an easy addition.

Overview

Publishervellum-ai
Repositoryvellum-assistant
Skill nameadd-mcp-integration
Stars
1.4K
Forks
208
Bundled files
Instructions only
LicenseMIT
Links
  • Markdown instructions

    A SKILL.md file the model loads on demand, so it only costs tokens when a request actually matches.

  • Works with any LLM

    AI skills are plain Markdown, not provider-specific code, so this works with GPT, Claude, Gemini, Grok, or a local model.

  • Self-contained

    Everything the model needs lives in the instructions — no extra files to sync.

  • Open source

    Published by vellum-ai on GitHub. Read the source before you install it.

Installation

Install the Add Mcp Integration AI skill in TypingMind to use it with any LLM, or drop it into another agent that reads SKILL.md.

1

Install in TypingMind

TypingMind installs a skill straight from its GitHub folder — it reads SKILL.md, bundles the resource files, and stores the result locally.

  1. Open the app and go to Plugins → Skills.
  2. Choose "Install from GitHub".
  3. Paste the skill folder URL below and confirm.
    https://github.com/vellum-ai/vellum-assistant/tree/main/.claude/skills/add-mcp-integration
  4. Enable the skill in any chat where you want it available.
Plugins → Skills → Add skill → From GitHub URL, then paste the folder URL and press Continue.
2

Install in another agent

Any agent that reads the Agent Skills format can use this skill — copy the folder into that agent's skills directory.

Claude Code — .claude/skills
git clone --depth 1 https://github.com/vellum-ai/vellum-assistant.git /tmp/vellum-assistant
mkdir -p .claude/skills
cp -r /tmp/vellum-assistant/.claude/skills/add-mcp-integration .claude/skills/add-mcp-integration
Restart Claude Code after copying so it picks up the new skill.

Use it in TypingMind

Enable Add Mcp Integration in any TypingMind chat and the model takes it from there. Its name and description sit in the system prompt, and the moment a request matches, the model loads the full instructions itself — you never invoke it by hand, and it costs no tokens until it is actually used.

The model loads Add Mcp Integration on its own as soon as a request matches it.

Works with any AI model

AI skills are plain Markdown instructions rather than provider-specific code, so Add Mcp Integration is not tied to the model it was written for. Install it once in TypingMind and use it with GPT-5, Claude, Gemini, Grok, DeepSeek, Mistral, Llama, or a local model you run yourself — all on your own API keys.

  • Loaded only when it is needed

    The system prompt carries just the name and description. The instructions are fetched on the first matching request, so an idle skill costs nothing.

  • Switch models mid-chat

    Because the skill is instructions rather than code, changing model does not break it — the next model reads the same SKILL.md.

Skill instructions

This is the SKILL.md content the model loads. Read it before installing — a skill is instructions your model will follow.

Add an MCP Integration

A catalog integration is a bundled local plugin whose only content is an mcp.json pointing at the vendor's remote server. The assistant ships it, the Integrations page lists it, and connecting it runs the MCP OAuth flow in assistant/src/mcp/mcp-oauth-provider.ts.

Scope: remote MCP servers in plugins/mcp-catalog/. For a full plugin that lives in another GitHub repo, see plugins/README.md § Marketplace.

Work in a worktree from origin/main. One provider per PR.

1. Research the server

From the vendor's MCP docs, record:

  • The remote URL. Prefer streamable-http (docs often say "type": "http").
  • The docs page URL. It is both homepage and documentationUrl.
  • A one-line summary of what the tools do.

Reject a server that only runs locally (npx, uvx, Docker). The catalog ships no stdio servers.

2. Check OAuth support

The catalog's one-step connect needs the server to support OAuth discovery and dynamic client registration (DCR), because a plugin cannot ship a client secret.

bash
URL=https://mcp.example.com/mcp
curl -sS -i -X POST "$URL" -H 'content-type: application/json' \
  -H 'accept: application/json, text/event-stream' \
  -d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2025-06-18","capabilities":{},"clientInfo":{"name":"probe","version":"0"}}}'

Expect 401 with WWW-Authenticate: Bearer ... resource_metadata="<url>". Fetch that URL and take authorization_servers[0] as the issuer. Then fetch the issuer's metadata from the first URL that answers, in the order the MCP SDK tries them (buildDiscoveryUrls in @modelcontextprotocol/sdk client/auth.js):

IssuerMetadata URLs, in order
https://auth.example.com/.well-known/oauth-authorization-server, then /.well-known/openid-configuration
https://auth.example.com/tenant1/.well-known/oauth-authorization-server/tenant1, then /.well-known/openid-configuration/tenant1, then /tenant1/.well-known/openid-configuration (all on the issuer host)

Check the metadata the way the SDK does:

FieldRule
registration_endpointMust be present. The SDK stops with "does not support dynamic client registration" without it.
code_challenge_methods_supportedIf present, must include S256. Absent is fine.
response_types_supportedMust include code.
grant_types_supportedShould include refresh_token. Without it, users sign in again each time the access token ends.

Do not judge token_endpoint_auth_methods_supported from the metadata. The registration in step 3 shows which method the server gives our public client.

Pick the setup mode:

  • The rules hold and step 3 succeeds: oauth. The normal case.
  • DCR works, but the vendor must allowlist our redirect URI first: manual. ramp is the example. Its setup.instructions tell the user what to ask the vendor for.
  • No registration_endpoint, or step 3 only returns a confidential client (a client_secret and a client_secret_* auth method): not a catalog addition. It needs a pre-registered client, which the MCP OAuth flow does not support. Stop and report this.

Also check whether the vendor already has a main OAuth provider in assistant/src/oauth/seed-providers.ts. If it does, set integration.oauthProvider so the Integrations page groups both connections (see calendly, linear, notion, todoist).

3. Test registration

Register a throwaway client with the same shape as McpOAuthProvider.clientMetadata:

bash
curl -sS -X POST <registration_endpoint> -H 'content-type: application/json' \
  -d '{"client_name":"Vellum Assistant (registration probe)","redirect_uris":["https://example.com/webhooks/oauth/callback"],"token_endpoint_auth_method":"none","grant_types":["authorization_code","refresh_token"],"response_types":["code"],"logo_uri":"https://www.vellum.ai/favicon.ico","software_version":"0.0.0"}'

A client_id with token_endpoint_auth_method: "none" in the response means the server registers public clients.

This probe does not prove the vendor accepts our real redirect URI. Each assistant resolves its own callback (resolveOauthCallbackUrl in assistant/src/inbound/oauth-callback-url.ts), and a vendor can restrict redirect hosts. If you know the callback URL of the assistant you will test with, register that instead of example.com. If the example.com probe is rejected for its redirect URI, retry with a real callback before you reject the server. Step 9 is the real test of the redirect.

4. Icon

Write plugins/mcp-catalog/<name>/icon.png: a PNG, 128x128, on an opaque white background so it reads on the dark theme. The inventory test requires at least 64x64.

Source, in order of preference:

  1. The vendor's official asset (apple touch icon, a large favicon, a press kit).
  2. Simple Icons, pinned to a commit URL, rasterized in the brand color.
bash
magick <source> -resize 128x128 -background white -alpha remove -alpha off -strip PNG24:plugins/mcp-catalog/<name>/icon.png

Look at the result with the Read tool before you continue.

Write ICON_ATTRIBUTION.md beside it in the same form as the others: source link, what was done to it, and "It is a trademark of and is used only to identify this integration." circleback, craft, and wix show the official-asset form. atlassian shows the Simple Icons form.

5. Package files

plugins/mcp-catalog/<name>/plugin.json:

json
{
  "$schema": "https://agent-plugins.org/schemas/1.0.0/plugin.schema.json",
  "name": "<name>",
  "version": "1.0.2",
  "description": "<Verb-first summary, under about 60 characters.>",
  "homepage": "<MCP docs URL>",
  "license": "MIT"
}

plugins/mcp-catalog/<name>/mcp.json. The server key must equal <name>:

json
{
  "$schema": "https://agent-plugins.org/schemas/1.0.0/mcp.schema.json",
  "mcpServers": {
    "<name>": { "type": "streamable-http", "url": "<remote URL>" }
  }
}

Append an entry to plugins/marketplace.json after the last local entry:

json
{
  "name": "<name>",
  "source": {
    "source": "local",
    "path": "plugins/mcp-catalog/<name>",
    "version": "<plugin.json version>"
  },
  "description": "<same as plugin.json>",
  "homepage": "<MCP docs URL>",
  "license": "MIT",
  "integration": {
    "kind": "mcp",
    "displayName": "<Vendor>",
    "documentationUrl": "<MCP docs URL>",
    "verifiedAt": "<today, YYYY-MM-DD>",
    "verification": "documentation-only",
    "setup": {
      "mode": "oauth",
      "instructions": "<One sentence: which account to sign in with.>"
    },
    "logo": "<name>-mcp.png",
    "category": "<slug>"
  }
}
  • category must be a slug from INTEGRATION_CATEGORIES in packages/service-contracts/src/integration-categories.ts.
  • homepage is the MCP docs URL, not the vendor's home page. Every catalog entry follows this.
  • No em dashes in any copy. Users read description and instructions.

6. Run the generators

From the repo root, in this order:

bash
node scripts/plugins/sync-local-plugin-icons.mjs              # web copy: clients/web/public/images/integrations/<name>-mcp.png
node scripts/plugins/generate-plugin-icons.mjs                # platform copy: plugins/assets/<name>/icon.png + plugins/plugin-icons.json
bun run meta/sync-bundled-copies.ts                           # bundled offline marketplace copy
bun run assistant/scripts/generate-bundled-plugin-packages.ts # bundled package map

git status must show changes for <name> only.

7. Update the inventory test

In scripts/plugins/__tests__/mcp-marketplace-inventory.test.ts:

  • Add <name> to EXPECTED_PROVIDERS.
  • Increase the provider count in the test title.
  • Increase the oauth count, or change the manual assertion.
  • Add to the oauthProvider map if you set one.

8. Run the CI checks

These are the steps of .github/workflows/pr-marketplace.yaml:

bash
node scripts/check-marketplace-prefix.mjs
bun run meta/sync-bundled-copies.ts --check
bun run assistant/scripts/generate-bundled-plugin-packages.ts --check
node scripts/plugins/generate-plugin-icons.mjs --check
node scripts/plugins/sync-local-plugin-icons.mjs --check
(cd scripts && bun test plugins/__tests__/bundled-plugin-packages.test.ts plugins/__tests__/mcp-marketplace-inventory.test.ts plugins/__tests__/add-plugin-icon.test.mjs plugins/__tests__/generate-plugin-icons.test.mjs plugins/__tests__/sync-local-plugin-icons.test.mjs)

9. Test end to end

Needs an account with the vendor. On an assistant built from the branch (see the cli-testing skill, or a platform preview):

  1. Open Integrations, find the tile, and connect. Finish the vendor sign-in.
  2. Confirm it shows as connected. assistant mcp list shows the server.
  3. Ask the assistant for something that calls one of the server's tools (for example, "list my projects") and confirm a real result.

If nobody can test this before the next release, add <name> to MCP_CATALOG_QA_INTEGRATION_NAMES in assistant/src/cli/lib/plugin-catalog-visibility.ts, and add the vendor to the mcp-catalog-qa-integrations description in meta/feature-flags/feature-flag-registry.json and clients/web/src/lib/feature-flags/feature-flag-registry.json. Then the entry stays hidden until it passes QA.

The flag key and default do not change, so this needs no platform Terraform change. Check its row in meta/feature-flags/PENDING_PLATFORM_PRS.md: while the flag is not provisioned, nobody can turn it on remotely, so QA uses a local flag override.

10. Open the PR

  • Title: feat(marketplace): add the <Vendor> MCP server to the integrations catalog.
  • In the body, show the OAuth evidence from steps 2 and 3, the icon source, and the step 8 commands.
  • Leave step 9 unchecked in the test plan if it was not done, and say why.

Frequently asked questions

What does the Add Mcp Integration AI skill do?

Add a vendor's remote MCP server to the bundled integrations catalog (plugins/mcp-catalog/<name>/ + plugins/marketplace.json) so users can connect it from the Integrations page. Covers vetting the server's OAuth support, the icon, the package files, the generators, the CI checks, and the PR. Use when asked to "add <vendor> MCP", "add <vendor> to integrations", or to check whether a vendor's MCP server is an easy addition.

Why use Add Mcp Integration on TypingMind?

Because you install it once and use it with any model. Add Mcp Integration is plain Markdown rather than provider-specific code, so the same skill runs on GPT-5, Claude, Gemini, Grok, or a local model — and you can switch model mid-chat without it breaking. TypingMind runs on your own API keys, so you pay providers directly instead of a per-seat subscription, and your skills and chats stay in your own storage.

How do I install Add Mcp Integration in TypingMind?

Open Plugins → Skills → Install from GitHub in TypingMind and paste https://github.com/vellum-ai/vellum-assistant/tree/main/.claude/skills/add-mcp-integration. TypingMind reads its SKILL.md and installs it as a skill you can enable per chat.

Which AI models can use Add Mcp Integration?

Any model you connect in TypingMind. AI skills are plain Markdown instructions rather than provider-specific code, so GPT, Claude, Gemini, Grok, and local models can all load this skill when a request matches it.

How many AI models can I use with Add Mcp Integration?

As many as you like. As long as a model supports skills, you can use Add Mcp Integration with it — GPT, Claude, Gemini, Grok, DeepSeek, Mistral, Llama and more — all on TypingMind with your own API keys.

Is the Add Mcp Integration AI skill free?

Yes. It is published on GitHub by vellum-ai under the MIT license. You only pay your own AI provider for the tokens you use.

What are AI skills?

An AI skill is a reusable instruction bundle that teaches an AI model how to do one specific task. It follows the open Agent Skills format: a SKILL.md file with a name and description, plus any scripts, templates or reference files the model may need. The model reads the instructions only when your request matches the skill, so an installed skill costs nothing until it is used.

How are AI skills different from plugins or MCP servers?

A plugin or MCP server gives a model new tools to call — code that runs somewhere and returns a result. An AI skill gives the model knowledge and process instead: how to approach a task, which steps to follow, what good output looks like. Skills are plain Markdown, so they need no server, no API key and no runtime, and they work with any model.

View all

Set up your own AI workspace now

Get notified about new features and future giveaways by subscribing to our newsletter 👇