MCP (Model Context Protocol) Integration

By default, Askimo answers questions using what it already knows from its training. MCP (Model Context Protocol) extends that by giving the AI the ability to act: to read your files, query a database, call an API, run code, and more, all from within your conversation.

Think of MCP servers as plugins that you connect to Askimo. Each server exposes a set of tools (actions the AI can perform). Once connected, the AI can choose the right tool at the right moment. For example, it can look up a live database record, fetch a web page, or list files in a folder, without you having to copy and paste anything manually.

You set up an MCP instance in Settings → MCP Servers by telling Askimo how to reach a server: what command to run or what URL to call. Askimo starts or connects to the server on demand and makes its tools available during your conversation.

Mẹo

You can connect multiple MCP servers at the same time, for example a database server, a file-system server, and a web-search server, and the AI will use whichever tools are relevant to your question.

How MCP Works in Askimo

graph TD
    I1[MCP Instance\ne.g. Personal GitHub]
    I2[MCP Instance\ne.g. Production DB]
    I3[MCP Instance\ne.g. Local Filesystem]

    I1 --> S1[MCP Server Process]
    I2 --> S2[MCP Server Process]
    I3 --> S3[MCP Server Process]

    S1 & S2 & S3 -->|tools exposed to AI| AI[Askimo AI]

Each MCP instance holds a complete set of connection details for one server: the command to run, environment variables, or a URL. Askimo starts or connects to the server when needed and exposes its tools to the AI in your conversation.

Adding an MCP Instance

MCP instances are managed in Settings → MCP Servers. From there you can add, edit, enable/disable, and delete instances that are available across your projects.

1. Click the Settings icon in the sidebar (or press ⌘ , on macOS / Ctrl , on Windows & Linux).
2. Select MCP Servers from the left menu.
3. Click the Add MCP Server button.
4. Fill in the instance details in the dialog that appears.
5. Click Test Connection to verify the server starts and returns tools.
6. Click Save to add the instance.

Instance Configuration

Name and Description

Field	Required	Description
Instance Name	Yes	A recognisable label for this connection (e.g. Personal GitHub, Production DB)
Description	No	A short note about what this instance does

Transport Type

Choose how Askimo connects to the MCP server:

Stdio

Askimo launches a local process and communicates over stdin/stdout.

Field	Description	Example
Command	The executable to run (can include the full command line)	npx -y @modelcontextprotocol/server-postgres
Arguments	Additional arguments passed after the command	--readonly
Working Directory	Optional directory the process starts in	/home/user/projects
Environment Variables	KEY=VALUE pairs (one per line) passed to the server process	DATABASE_URL=postgresql://localhost/mydb

HTTP

Askimo connects to a remote or locally running HTTP MCP endpoint.

Field	Description	Example
URL	Full URL of the MCP server endpoint	https://my-mcp-server.example.com/mcp
Headers	Header-Name=value pairs (one per line)	Authorization=Bearer mytoken
Timeout (ms)	Request timeout in milliseconds	60000

Testing the Connection

Click Test Connection before saving to verify everything is set up correctly. Askimo will:

1. Start the server process (Stdio) or contact the URL (HTTP) using the configuration you entered.
2. Fetch the list of available tools from the server.
3. Display the tools found and show a ✅ success or ❌ failure indicator.

You can save the instance even without testing first, but testing is recommended so you know the tools are reachable.

Mẹo

For Stdio servers, make sure the binary is installed and on your PATH (e.g. run npx --version or uvx --version in a terminal) before clicking Test Connection.

Include MCP Tools in Your Conversation

Once you have set up MCP instances, you can control which tools are active in any chat conversation using the Tools selector in the chat input bar.

Clicking the Tools button opens a panel listing all available tool sources. You can toggle each one on or off for the current conversation:

• MCP tools (your connected MCP instances) are enabled by default. The AI will use them automatically when relevant to your question.
• Built-in Askimo tools are disabled by default. You can turn them on when you want Askimo to use its own internal capabilities alongside your MCP servers.

Changes to the tool selection apply to the current conversation only and do not affect your global MCP instance settings.

Security: Secret Values

Askimo automatically detects sensitive field values based on their name. Any environment variable or header whose name contains key, secret, token, password, or credential is treated as a secret:

• Its value is never written to disk in plain text.
• It is stored in your operating system keychain and retrieved at runtime.

This means a GITHUB_TOKEN or DATABASE_PASSWORD environment variable is handled securely without any extra steps on your part.

Chú ý

Keychain entries are not removed when you delete an MCP instance or uninstall Askimo. If you need to remove them, do so manually via Keychain Access (macOS) or the system credential store (Windows / Linux).

Managing Existing Instances

In Settings → MCP Servers you can:

Action	How
View available tools	Click the tools icon on an instance card
Enable / disable	Toggle the instance on or off without deleting it
Edit	Click the edit icon to update the configuration
Delete	Click the delete icon and confirm

Tool Categories

Askimo classifies MCP tools into categories that help the AI decide when and how to use them:

Category	Description
DATABASE	SQL/NoSQL queries and data operations
NETWORK	HTTP requests and API calls
FILE_READ	Reading files and directories
FILE_WRITE	Writing, creating, or deleting files
VISUALIZE	Charts, graphs, and data visualisation
EXECUTE	Running code or shell commands
SEARCH	Full-text and semantic search
TRANSFORM	Data conversion and processing
VERSION_CONTROL	Git and VCS operations
COMMUNICATION	Email, messaging, notifications
MONITORING	Logging and observability
OTHER	Unclassified tools

Human Guardrails: Approval Policies

Some tools are safe to let the AI run freely (reading a file, searching the web); others can change or delete data and deserve a second look before they run. Askimo lets you set a per-tool approval policy so you decide where that line is, instead of trusting a blanket “always ask” or “never ask” rule for every tool.

Each tool resolves to one of two effective policies, shown in the UI:

Effective Policy	Behaviour
Require Approval	Askimo pauses before executing the tool and asks you to confirm.
Always Auto	The tool runs immediately, without a confirmation prompt.

Under the hood, every tool also has a category (see Tool Categories) that implies a sensible default: tools that write files or execute code (FILE_WRITE, EXECUTE) default to Require Approval, while read-only and lower-risk categories default to Always Auto. You never need to think about this default directly — the Tools panel always shows you the concrete, effective outcome for each tool, not an abstract “default” state.

You can override the default on any tool at any time:

• Flip a normally safe tool to Require Approval if you want an extra guardrail around it (e.g. a search tool that also happens to hit a paid API).
• Flip a normally sensitive tool to Always Auto once you trust it, so the AI can act without interrupting your flow.

This gives you fine-grained human-in-the-loop control: the AI can move fast on everyday, low-risk actions, while sensitive operations always pause for your explicit go-ahead, unless you’ve decided otherwise.

Enabling or Disabling Individual Tools

A single MCP server can expose anywhere from a handful to dozens of tools. By default Askimo makes every discovered tool available to the AI, but you can disable individual tools you don’t need — per MCP instance, directly from the Tools panel or the instance’s tool list.

Disabling specific tools matters for two practical reasons:

• Token cost. Every enabled tool’s definition is sent to the model with each request. A server that exposes many tools (e.g. a full database or cloud-provider MCP) can add a meaningful number of tokens to every message, even when you only ever use two or three of them.
• Tool selection accuracy. When the AI has to choose among a large number of similarly-named or overlapping tools, it can struggle to pick the right one for the task, an effect sometimes called “tool confusion.” Narrowing the toolset to only what you actually use reduces ambiguity and improves the AI’s accuracy.

If you know your workflow never needs a particular tool (e.g. a delete_table tool on a database server you only ever query), disable it. The tool stays discoverable in the UI so you can re-enable it later, but it’s excluded from what’s sent to the model, keeping requests smaller and tool selection sharper.

Troubleshooting

“Failed to connect to MCP server”

• Verify that the binary required by the command is installed and on your PATH (e.g. run npx --version or uvx --version in a terminal).
• Double-check any environment variable names and values, especially tokens or passwords.
• For HTTP servers, confirm the server is running and the URL is reachable.

Tools list is empty after a successful connection

• Some MCP servers return an empty tool list if authentication fails silently. Double-check any token or API key values in your environment variables or headers.
• Try editing the instance and clicking Test Connection again after correcting the values.

Security Notes

• Secret parameters (environment variables or headers whose names contain key, secret, token, password, or credential) are stored in your operating system keychain and never written to configuration files.
• MCP server processes inherit Askimo’s environment. Avoid granting file-write tools access to sensitive directories.
• For HTTP MCP servers, prefer HTTPS endpoints.

---

Learn More

• MCP Integration Guide (blog) - Practical walkthrough: connect GitHub, Time Server, and real-world tool examples with screenshots
• Askimo Features: MCP Tools - Visual overview of what MCP enables in the desktop app