# MCP Server (/es/data-api/guides/ai-tools/mcp)

Connect AI tools directly to Fiskil documentation using Model Context Protocol



import { Callout } from 'fumadocs-ui/components/callout';

The Fiskil MCP server gives your AI assistant direct access to Fiskil documentation, API references, and code examples. It works with any editor or tool that supports [Model Context Protocol](https://modelcontextprotocol.io/) — including Codex, Claude Code, Cursor, GitHub Copilot, and Gemini CLI.

**What your AI assistant can do once connected:**

* Full-text search across all documentation
* Retrieve complete page content including API reference details
* Browse the documentation tree by product and section
* Look up API endpoint details with request/response schemas
* Get copy-paste code examples in TypeScript, Python, JavaScript, and cURL
* Follow guided prompts for onboarding and integration
* Optionally report documentation gaps or MCP issues to Fiskil

Server URL [#server-url]

```
https://docs.fiskil.com/api/ai/mcp
```

Setup [#setup]

Use the **Streamable HTTP** transport. No Fiskil API key, bearer token or OAuth login is required for this public documentation server. Merge the examples below into existing configuration files.

Codex [#codex]

Run this in your terminal with the Codex CLI installed:

```bash
codex mcp add fiskil --url https://docs.fiskil.com/api/ai/mcp
codex mcp list
codex mcp get fiskil
```

The CLI and Codex IDE extension share MCP configuration. You can also add the server to `~/.codex/config.toml`:

```toml title="~/.codex/config.toml"
[mcp_servers.fiskil]
url = "https://docs.fiskil.com/api/ai/mcp"
```

Start a new Codex session after configuring it. See the [official Codex setup instructions](https://developers.openai.com/learn/docs-mcp).

Claude Code [#claude-code]

Run from your project directory:

```bash
claude mcp add --transport http --scope project fiskil https://docs.fiskil.com/api/ai/mcp
claude mcp get fiskil
```

This writes a shared project configuration to `.mcp.json`. Review the project server when Claude Code asks to approve it. Use `--scope user` instead for a personal configuration across projects. Without `--scope`, Claude Code uses local scope: private to you in the current project.

Alternatively, merge this entry into your project-root `.mcp.json`:

```json title=".mcp.json"
{
  "mcpServers": {
    "fiskil": {
      "type": "http",
      "url": "https://docs.fiskil.com/api/ai/mcp"
    }
  }
}
```

Run `/mcp` inside Claude Code to check the connection. Prefer the CLI for user-level changes rather than replacing `~/.claude.json`. See [Claude Code's MCP documentation](https://code.claude.com/docs/en/mcp).

Cursor [#cursor]

Merge this into `.cursor/mcp.json` for the project, or `~/.cursor/mcp.json` for all projects:

```json title=".cursor/mcp.json"
{
  "mcpServers": {
    "fiskil": {
      "url": "https://docs.fiskil.com/api/ai/mcp"
    }
  }
}
```

Check the server in Cursor's MCP settings. The Cursor CLI uses the same configuration:

```bash
agent mcp list
agent mcp list-tools fiskil
```

See [Cursor's MCP documentation](https://cursor.com/docs/mcp).

GitHub Copilot in VS Code [#github-copilot-in-vs-code]

In a Copilot session, run **MCP: Add Server** from the Command Palette. Select HTTP, enter the server URL and use `fiskil` as the name. Choose **Copilot Global** for your personal configuration, or the workspace-root `.mcp.json` option when available.

For current Copilot configurations, use the portable `mcpServers` format:

```json title=".mcp.json"
{
  "mcpServers": {
    "fiskil": {
      "type": "http",
      "url": "https://docs.fiskil.com/api/ai/mcp",
      "tools": ["*"]
    }
  }
}
```

For personal use, merge the same entry into `~/.copilot/mcp-config.json` (or the location selected by `COPILOT_HOME`). This also supports Copilot CLI on the same host. Existing editor-managed `.vscode/mcp.json` files use a different `servers` format; do not paste this example into that file.

Open **Chat: Open Customizations → MCP Servers** in the intended session and verify a tool call. See the [current Copilot configuration reference](https://code.visualstudio.com/docs/agents/reference/mcp-configuration).

Gemini CLI [#gemini-cli]

Run from your project directory:

```bash
gemini mcp add --transport http --scope project fiskil https://docs.fiskil.com/api/ai/mcp
gemini mcp list
```

Use `--scope user` instead to configure it across projects. Check tools with `/mcp` in a Gemini CLI session. See [Gemini CLI's MCP documentation](https://geminicli.com/docs/tools/mcp-server/).

Other MCP clients [#other-mcp-clients]

For Windsurf or another client, add `https://docs.fiskil.com/api/ai/mcp` using its remote HTTP MCP settings. Configuration keys and file locations vary by client; follow that client's current instructions.

Verify the connection [#verify-the-connection]

Registration only saves configuration. In your assistant, ask:

```text
Use Fiskil's search_docs tool to find the Link SDK integration guide, then
use get_page to read it. Return the page URL and the SDK success-result shape.
```

The assistant should make tool calls and report `consents: string[]` with an optional `redirectURL`. If it answers from memory, explicitly ask it to use the `fiskil` MCP server.

***

Tools [#tools]

The MCP server provides six documentation tools and an optional feedback tool.

search_docs [#search_docs]

Full-text search across Fiskil documentation. Searches page titles, headings, and body content. Returns matching pages with relevant text snippets and direct URLs.

| Parameter | Type   | Required | Default | Description                                            |
| --------- | ------ | -------- | ------- | ------------------------------------------------------ |
| `query`   | string | Yes      | —       | Search query to find relevant documentation            |
| `product` | string | No       | `all`   | Filter by product: `all`, `data-api`, or `data-holder` |
| `limit`   | number | No       | `10`    | Maximum number of results (1–20)                       |

**Example prompts:**

> "Search Fiskil docs for webhook signature verification"
>
> "Search for authentication in the Data API docs, limit to 5 results"

Each result includes a `fullUrl` field with the absolute URL you can open directly in your browser.

***

get_page [#get_page]

Retrieve the full markdown content of a specific documentation page. For API reference pages, this includes the HTTP method, path, parameters, request/response schemas, and examples.

| Parameter | Type   | Required | Description                                                               |
| --------- | ------ | -------- | ------------------------------------------------------------------------- |
| `path`    | string | Yes      | URL path of the page, e.g. `/data-api/guides/getting-started/quick-start` |

**Example prompts:**

> "Get the full content of the Fiskil authentication guide"
>
> "Fetch the page at /data-api/guides/core-concepts/webhooks"

<Callout type="info">
  If the page is not found, the response will suggest using `search_docs` or `list_pages` to discover valid paths.
</Callout>

***

list_pages [#list_pages]

Browse the documentation tree. Returns all available pages grouped by product and section. Use this to discover what documentation exists before searching or fetching specific pages.

| Parameter | Type   | Required | Default | Description                                                         |
| --------- | ------ | -------- | ------- | ------------------------------------------------------------------- |
| `product` | string | No       | `all`   | Filter by product: `all`, `data-api`, or `data-holder`              |
| `section` | string | No       | `all`   | Filter by section: `all`, `guides`, `api-reference`, or `changelog` |

**Example prompts:**

> "List all Fiskil Data API guide pages"
>
> "Show me the API reference pages for Data API"

***

list_api_endpoints [#list_api_endpoints]

List API endpoints from the Fiskil OpenAPI reference. Returns the HTTP method, path, summary, operationId, and tags for each endpoint.

| Parameter  | Type   | Required | Default | Description                                                                                                 |
| ---------- | ------ | -------- | ------- | ----------------------------------------------------------------------------------------------------------- |
| `product`  | string | No       | `all`   | Filter by product: `all`, `data-api`, or `data-holder`                                                      |
| `category` | string | No       | —       | Filter by category/tag (case-insensitive), e.g. `Banking`, `Energy`, `auth`, `Common`, `cdr`, `fdx`, `psd2` |

**Example prompts:**

> "List all Fiskil banking API endpoints"
>
> "Show me the energy endpoints in the Data API"

<Callout type="info">
  Data Holder endpoints (CDR, FDX, PSD2) are currently in preview. If Data Holder is not yet enabled for your environment, the tool will let you know and suggest contacting Fiskil for early access.
</Callout>

***

get_api_endpoint_details [#get_api_endpoint_details]

Look up detailed information for a specific API endpoint. Returns parameters, request/response schemas with field-level descriptions, examples, and a ready-to-use cURL snippet.

You can look up an endpoint in two ways:

1. **By method + path** — e.g. `GET /v1/banking/accounts`
2. **By operationId** — e.g. `getBankingTransactionsV1` or `create-auth-session`

| Parameter     | Type   | Required    | Default    | Description                                                                               |
| ------------- | ------ | ----------- | ---------- | ----------------------------------------------------------------------------------------- |
| `method`      | string | Conditional | —          | HTTP method: `GET`, `POST`, `PUT`, `PATCH`, or `DELETE`. Required when using path lookup. |
| `path`        | string | Conditional | —          | API path, e.g. `/v1/banking/accounts`; copy the path from `list_api_endpoints`            |
| `operationId` | string | Conditional | —          | Operation ID or slug, e.g. `getBankingTransactionsV1`. Alternative to method + path.      |
| `product`     | string | No          | `data-api` | Which product to search: `data-api` or `data-holder`                                      |

<Callout type="info">
  You must provide **either** `method` + `path` **or** `operationId`. Copy the path or operation ID from `list_api_endpoints`. Lookup paths reflect the OpenAPI files and can differ from the public request URL namespace. Confirm the chosen API version in the [API reference](/data-api/api-reference) and [versioning guide](/data-api/guides/core-concepts/versioning).
</Callout>

**Example prompts:**

> "Get details for GET /v1/banking/accounts from Fiskil"
>
> "Look up the getBankingTransactionsV1 endpoint"
>
> "Look up the create-auth-session operation"

If the endpoint is not found, the tool suggests similar endpoints you might have meant.

***

get_code_examples [#get_code_examples]

Get starter code examples for common Fiskil integration patterns. Available in TypeScript, Python, JavaScript, and cURL. Not every topic has every language variant.

Banking and energy examples fetch one page using `end_user_id` and `X-Fiskil-Version: v3`. Follow the pagination links for additional pages. Webhook examples demonstrate verification and event dispatch; implement durable deduplication and your application handlers before using them in production.

| Parameter  | Type   | Required | Default       | Description                                                                                                   |
| ---------- | ------ | -------- | ------------- | ------------------------------------------------------------------------------------------------------------- |
| `topic`    | string | Yes      | —             | Integration topic: `auth`, `end-users`, `auth-session`, `link-sdk`, `webhooks`, `banking`, `energy`, or `all` |
| `language` | string | No       | all languages | Programming language: `typescript`, `python`, `javascript`, or `curl`                                         |

**Available topics:**

| Topic          | What it covers                                          |
| -------------- | ------------------------------------------------------- |
| `auth`         | Authenticate with client credentials (`POST /v1/token`) |
| `end-users`    | Create end users                                        |
| `auth-session` | Create auth sessions for the Link SDK                   |
| `link-sdk`     | Embed the Link SDK in React/JavaScript                  |
| `webhooks`     | Webhook verification and event handling                 |
| `banking`      | Fetch banking data (accounts, transactions, balances)   |
| `energy`       | Fetch energy data (usage, billing, service points)      |
| `all`          | All of the above                                        |

**Example prompts:**

> "Show me a TypeScript example for Fiskil authentication"
>
> "Get Python code examples for handling webhooks"
>
> "Show all Fiskil banking code examples"

***

submit_feedback [#submit_feedback]

Send optional feedback about Fiskil documentation or MCP tools. Useful feedback describes a missing page, an incorrect or unclear example, a tool issue, or a suggested improvement. Submit once per distinct issue after using the documentation.

| Parameter               | Type    | Required | Description                                                                               |
| ----------------------- | ------- | -------- | ----------------------------------------------------------------------------------------- |
| `feedback_type`         | string  | Yes      | `missing_capability`, `issue`, `praise`, or `other`                                       |
| `summary`               | string  | Yes      | A concise description of the observation                                                  |
| `details`               | string  | No       | Relevant context, without credentials or personal data                                    |
| `sentiment`             | string  | No       | `positive`, `neutral`, `negative`, or `mixed`                                             |
| `tool_name`             | string  | No       | The tool this feedback concerns                                                           |
| `task_completed`        | boolean | No       | Whether the agent completed its task                                                      |
| `friction_points`       | string  | No       | What made the task difficult                                                              |
| `suggested_improvement` | string  | No       | A concrete change that would help                                                         |
| `product`               | string  | No       | `data-api`, `data-holder`, or `general`                                                   |
| `category`              | string  | No       | `missing_docs`, `incorrect_docs`, `unclear_docs`, `tool_error`, `suggestion`, or `praise` |
| `page_path`             | string  | No       | A relative documentation path, without a query string or fragment                         |

For example:

```json
{
  "feedback_type": "issue",
  "summary": "The example does not explain how to handle an expired session",
  "suggested_improvement": "Add an example showing session recreation and retry",
  "product": "data-api",
  "category": "unclear_docs",
  "tool_name": "get_page",
  "page_path": "/data-api/guides/getting-started/quick-start",
  "task_completed": false
}
```

Feedback records the agent's observation, rather than a rating from the developer. Keep each text field under 2,000 characters. Delivery is best effort. For help with a live integration, use [Fiskil support](mailto:support@fiskil.com).

The server advertises this tool when analytics is configured. When a response includes a `conversation_id`, pass it unchanged in subsequent tool calls to associate feedback with the same documentation session. It is optional; existing calls continue to work without it.

***

Prompts [#prompts]

The MCP server includes two guided prompts that provide structured workflows. Your AI assistant can use these to give you step-by-step guidance.

fiskil_getting_started [#fiskil_getting_started]

A recommended workflow for new developers integrating with Fiskil APIs. No parameters required.

Walks you through a 6-step onboarding process:

1. **Authenticate** — get an access token with client credentials
2. **Create an End User** — register the user who will link their accounts
3. **Create an Auth Session** — generate a session for the Link SDK
4. **Embed the Link SDK** — install and launch the consent flow in your app
5. **Set Up Webhooks** — listen for data sync events with signature verification
6. **Fetch Data** — use the required `end_user_id` and any account filters from the endpoint reference

**Example prompt:**

> "Use the Fiskil getting started guide to walk me through integration"

fiskil_integration_guide [#fiskil_integration_guide]

A tailored integration guide filtered by your use case. Returns relevant API endpoints, webhook events, and implementation guidance.

| Parameter  | Type   | Required | Description                                              |
| ---------- | ------ | -------- | -------------------------------------------------------- |
| `use_case` | string | Yes      | The data domain you need: `banking`, `energy`, or `both` |

**What's included by use case:**

* **banking** — Accounts, Transactions, Balances, Payees, Direct Debits, Scheduled Payments endpoints + `banking.transactions.sync.completed` webhook
* **energy** — Energy Accounts, Usage, Billing, Service Points, DER, Concessions endpoints + `energy.usage.sync.completed` webhook
* **both** — All of the above

**Example prompt:**

> "Give me the Fiskil integration guide for banking"

***

Resources [#resources]

docs://overview [#docsoverview]

The MCP server exposes a documentation overview resource at `docs://overview`. This provides a summary of:

* Available products (Data API, Data Holder) with page counts
* All available MCP tools and prompts
* Quick links to key documentation pages
* Links to LLM resources

***

Usage Tips [#usage-tips]

1. **Start with search** — ask your AI to search for topics rather than guessing page URLs
2. **Browse first** — use `list_pages` to discover what documentation is available
3. **Get full pages** — once you find relevant results, use `get_page` for the complete content
4. **Look up endpoints** — use `list_api_endpoints` to find endpoints, then `get_api_endpoint_details` for schemas and examples
5. **Grab code snippets** — use `get_code_examples` to get working starter code in your language
6. **Follow guided prompts** — use the getting started or integration guide prompts for a structured walkthrough
7. **Start a focused task** — use the short prompt in [Build with AI](/data-api/guides/ai-tools)

***

Troubleshooting [#troubleshooting]

Connection Issues [#connection-issues]

If your AI tool can't connect to the MCP server:

1. Verify the server URL is exactly `https://docs.fiskil.com/api/ai/mcp`
2. Check that your network allows outbound HTTPS connections
3. Restart your AI application after making configuration changes
4. Confirm that your client is using Streamable HTTP and its own configuration format
5. **Verify registration** — use your client's MCP list/status command, then run the test prompt above. A configured server is not necessarily connected.

No Results [#no-results]

If searches return no results:

1. Try broader or different search terms
2. Use the `product` filter to narrow scope (e.g., `data-api`)
3. Use `list_pages` to browse available documentation
4. Check [`/llms.txt`](https://docs.fiskil.com/llms.txt) for a list of all available pages

Rate Limiting [#rate-limiting]

If a request receives `429 Too Many Requests`, follow `Retry-After` when present and back off before retrying. Data API rate limits are separate from the documentation connection.

***

LLM Resources [#llm-resources]

In addition to MCP, the following resources are available for AI tools:

| Resource        | URL                                                     | Description                                                              |
| --------------- | ------------------------------------------------------- | ------------------------------------------------------------------------ |
| `llms.txt`      | [/llms.txt](https://docs.fiskil.com/llms.txt)           | Lightweight index of all documentation pages                             |
| `llms-full.txt` | [/llms-full.txt](https://docs.fiskil.com/llms-full.txt) | Full documentation content in a single file                              |
| `skill.md`      | [/skill.md](https://docs.fiskil.com/skill.md)           | Capabilities and documentation reference; not an installable agent skill |

***

Security [#security]

The documentation tools provide **read-only** access to public documentation and require no authentication. The optional `submit_feedback` tool sends feedback to Fiskil.

Fiskil uses PostHog to record MCP usage metadata, including tool names, timing, errors, client identity when available, and anonymous conversation identifiers. Routine tool arguments and response bodies are excluded. Explicit feedback text is recorded with length limits and best-effort redaction. Never include API credentials, tokens, personal details, customer data, or conversation transcripts in feedback.
